Citation
Detecting falls in residents with dementia in a memory care facility using a real-time wireless fall detection device

Material Information

Title:
Detecting falls in residents with dementia in a memory care facility using a real-time wireless fall detection device a pilot study
Creator:
Christensen, Mackenzie L. ( author )
Language:
English
Physical Description:
1 electronic file (167 pages) : ;

Subjects

Subjects / Keywords:
Falls (Accidents) ( lcsh )
Falls (Accidents) -- Prevention ( lcsh )
Genre:
bibliography ( marcgt )
theses ( marcgt )
non-fiction ( marcgt )

Notes

Review:
The collaboration between the Neurocognitive Technology for Aging Lab and Anthem Memory Care identified a significant need for a real-time, comfortable, non-obtrusive fall detection device that can be worn by individuals with dementia. There is currently no fall detection technology built specifically for this population despite the fact that they are nearly twice as likely to fall compared to someone without dementia. A fall detection device was designed to fill this need. It consisted of a small microcontroller and a 10-Degree of Freedom IMU sensor. The device detects the direction of a fall, and sends a notification to a smartphone application. Eighteen young healthy individuals tested the fall detection device. Each subject completed 9 different falls into a matt and 4 activities of daily living (ADL). It resulted in a specificity, sensitivity, and accuracy of 100%, 95%, and 96% respectively and was 66% accurate when identifying the direction of the fall. Once this was complete, 9 residents of Anthem’s Chelsea Place Memory Care wore a mock fall detection device for 3 days, and then wore the real fall detection device for another 3 days. This was completed to determine that the device was comfortable and unobtrusive to the daily lives of the residents.
Thesis:
Thesis (M.S.)--University of Colorado Denver
Bibliography:
Includes bibliographical references.
System Details:
System requirements: Adobe Reader.
Statement of Responsibility:
by Mackensie L. Christensen.

Record Information

Source Institution:
University of Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
on10030 ( NOTIS )
1003046301 ( OCLC )
on1003046301

Downloads

This item has the following downloads:


Full Text
DETECTING FALLS IN RESIDENTS WITH DEMENTIA IN A MEMORY CARE
FACILITY USING A REAL-TIME WIRELESS FALL DETECTION DEVICE: A
PILOT STUDY
by
MACKENZIE L. CHRISTENSEN Bachelor of Science, Rose-Hulman Institute of Technology, 2015
A thesis submitted to the Faculty of the Graduate School of the University of Colorado in partial fulfillment of the requirements for the degree of Master of Science Bioengineering Program
2017


This thesis for the Master of Science degree by Mackenzie L. Christensen has been approved for the Bioengineering Program by
Cathy Bodine, Chair Levin Sliker Steven Lammers
Date: May 13, 2017
11


Christensen, Mackenzie L. (M.S., Bioengineering Program)
Detecting Falls in Residents with Dementia in a Memory Care Facility Using a Real-Time Wireless Fall Detection Device: A Pilot Study Thesis directed by Associate Professor Cathy Bodine
ABSTRACT
The collaboration between the Neurocognitive Technology for Aging Lab and Anthem Memory Care identified a significant need for a real-time, comfortable, non-obtrusive fall detection device that can be worn by individuals with dementia. There is currently no fall detection technology built specifically for this population despite the fact that they are nearly twice as likely to fall compared to someone without dementia. A fall detection device was designed to fill this need. It consisted of a small microcontroller and a 10-Degree of Freedom IMU sensor. The device detects the direction of a fall, and sends a notification to a smartphone application. Eighteen young healthy individuals tested the fall detection device. Each subject completed 9 different falls into a matt and 4 activities of daily living (ADL). It resulted in a specificity, sensitivity, and accuracy of 100%, 95%, and 96% respectively and was 66% accurate when identifying the direction of the fall. Once this was complete, 9 residents of Anthems Chelsea Place Memory Care wore a mock fall detection device for 3 days, and then wore the real fall detection device for another 3 days. This was completed to determine that the device was comfortable and unobtrusive to the daily lives of the residents.
The form and content of this abstract are approved. I recommend its publication.
Approved: Cathy Bodine
ill


ACKNOWLEDGEMENTS
Thank you so much to the many inspiring professors I have had, my committee members, friends, and my amazing parents for helping, supporting, and encouraging me throughout my educational endeavors. I could not have made it here without you!
IV


TABLE OF CONTENTS
CHAPTER
I. INTRODUCTION........................................................1
Increasing Population of People over 65 with Dementia...........1
Current Fall Detection Technology Used..........................3
Video Cameras.......................................................3
Smartphones.........................................................5
Microphones.........................................................7
Floor Sensors.......................................................8
Doppler Radar......................................................10
Impact of Technology to Detect or Predict Falls................18
Partnership with Anthem Memory Care............................19
II. SPECIFIC AIMS......................................................20
III. MATERIALS AM) METHODS..............................................21
Fall Detection System Development..............................21
Young Healthy Volunteer Study..................................30
Mock Device Study..............................................34
Final Device Testing...........................................38
IV. RESULTS.............................................................44
Fall Detection System Development..............................44
Young Healthy Volunteer Study..................................45
Mock Device Study..............................................60
Final Device Testing...........................................63
v


V. DISCUSSION AND FUTURE WORK
66
Fall Detection System Development..................................66
Young Healthy Volunteer Study......................................67
Mock Device and Final Device Study.................................69
Future Fall Detection System Studies...............................70
VI. CONCLUSION..............................................................73
REFERENCES..................................................................74
APPENDIX
A. Arduino Fall Detection Code....................................78
B. Python Fall Data Collection Code...............................98
C. Arduino Code to Test Battery Life Constantly Connected to Wifi.100
D. Arduino Code to Test Battery Life Only Connecting to WiFi When Fall
Triggered.........................................................106
E. Python Code Collecting Sensor Data............................112
F. Python Code Collecting Battery Data...........................114
G. Fall Backward End on Left Shoulder Graphs for Young............116
H. Fall Forward End on Left Shoulder Graphs.......................118
I. Lateral Left Fall Graphs......................................120
J. Fall Forward End on Right Shoulder Graphs.......................124
K. Fall Backwards End on Right Shoulder Graphs....................127
L. Lateral Fall Right Graphs.....................................129
M. Lateral Fall Left End on Stomach Graphs........................134
N. Lateral Right Fall End on Stomach Graphs......................139
vi


O. Fall Forward Graphs............................................144
P. Fall Backwards, Land in Sitting Position, End on Back Graphs...149
Q. Fall Backwards Graphs..........................................154
R. System Usability Survey........................................159
vii


CHAPTER I
INTRODUCTION
Increasing Population of People over 65 with Dementia
According to the United States Census Bureau, In 2050, the population aged 65 and over is projected to be 83.7 million, almost double its estimated population of 43.1 million in 2012, (1). This projection will be one of the biggest challenges facing the nation, and the world, in coming years. One of the most concerning issues for this aging population is that each individual will have their own medical concerns that will have to be addressed, and paid for. One of the most prevalent diseases in the aging population is dementia (2). According to the World Health Organization The number of people living with dementia worldwide is currently estimated at 47.5 million and is projected to increase to 75.6 million by 2030 (3).
Dementia is the decline of memory or other thinking skills such as memory loss, judgment, language, complex motor skills, and other intellectual function due to permanent damage or death of the brains nerve cells or neurons (3). These symptoms are often severe enough to reduce a person's ability to perform everyday activities (3). There are many different types of dementia such as lewy body, vascular, frontotemporal, and Alzheimers, which is the most common. Alzheimers is a debilitating disease and is the fifthdeading cause of death in the United States for those age 65 and older (3). It is the only top 10 cause of death in America today that cannot be prevented, cured, or slowed (3). The only treatment options available are monitoring of blood pressure, healthy diet, regular exercise, cognitive stimulation therapy, and social networks (4). Another treatment is prescription drugs. Of the approved drugs, such as cholinesterase and
1


memantine, it has been shown that the overall risks do not warrant using them, and that they do not slow the progression of dementia (5). Recent research has also shown that the best way to treat individuals with dementia is with customized care using creative management of physical therapies, environmental factors, and behavioral therapy (6).
Increased Risk of Falling Due to Dementia
Most treatments for persons with dementia are aimed at delaying the onset of common symptoms. However, increased probability of falling is one symptom that has few preventative treatment options. Physical therapy and environmental preventative measures can reduce the risk of falling (7). This can include something as simple as short daily exercises of walking (7). Other common measures are de-cluttering rooms, increasing lighting, and adding assistive devices, such as canes, walkers, wheelchairs, and fall pads (8). Despite these efforts, the unadjusted fall rate was 4.05 per year for residents in a nursing home diagnosed with dementia, compared with 2.33 falls per year for residents without dementia (9). This is mostly due to symptoms such as impaired judgment, gait instability, visual-spatial perception, and a decreased ability to recognize and avoid hazards (10).
For individuals with dementia who experience frequent falls there is a concern whether anyone will be present when a fall happens. The results of a fall can be as minor as a bruise or as serious as a traumatic brain injury, or even death (11). Every elderly individual who falls develops a fear of falling again and this fear leads to the individual falling more often (12). Researchers call this phenomenon space phobia or a persons loss of confidence in balance and walking. Community-based epidemiologic studies have found that 21-61% of elderly people experience some degree of fear of falling even if
2


they have never previously fallen. Furthermore, of the elderly individuals who are afraid of falling 70% say that they avoid certain activities due to this fear. This can often lead to depression and anxiety (13).
Current Fall Detection Technology Used
Given the increased fall rate in people with dementia researchers are working to develop reliable technologies to detect falls. Currently, they can be classified into a few different groupings: Video Cameras, Smartphones, Microphones, Floor Sensors, Doppler, and accelerometers/gyroscope/barometer sensor combinations. In the following paragraphs each of these will be discussed.
Video Cameras
Significant research has been completed on the use of video cameras to detect falls due to their ability to monitor an individual without being obtrusive to their daily life. They are also very accurate, as can be seen in Auvinet, Multon, Saint-Arnaud, Rousseau, & Meunier, 2011 (14) andBelshaw, Taati, Snoek& Mihailidis, 2011 (15).
Belshaw, et al., 2011 utilized an artificially intelligent camera-based system that automatically detects if a person within the field-of-view has fallen. The system uses a consumer grade camera with a wide-angle lens. A machine learning technique allowes the system to classify whether a person has fallen or not at a high accuracy rate. It takes into account lighting, environment and the presence of moving objects. The testing resulted in a true positive rate of 92% and a false positive rate of 5%. The large advancement in this study was that the system was able to handle multiple moving objects in a room and determine when one of those objects fell. It also accounted for large lighting changes throughout the day (15).
3


Auvinet et als., 2011 study provided another advancement by utilizing a new method of reconstructing the 3-D shape of people. Falls were detected by analyzing the volume distribution along the vertical axis, and an alarm triggered when the major part of this distribution was abnormally near the floor during a predefined period of time. This method was tested with videos of healthy subjects who performed 24 realistic scenarios showing 22 fall events and 24 activities of daily living. A 99.7% sensitivity and specificity was achieved with four cameras or more. The sensitivity decreased to 80.6% when they decreased the number of cameras to three. The researchers concluded that this study compares well with other literature, but acknowledge that falls that end up on a couch or chair could be missed due to the body not being close to the floor. They also acknowledge that the only way to use this system in a multi-room setting would require setting up 4 to 6 cameras in each room (14).
Despite the advantages of using camera systems, there were quite a few problems with the technology. The most important problem was that it required the recording of people, which means their privacy was being violated. This is not acceptable in a residential memory care community that values their residents anonymity. Sherwin & Winsby (2010) discuss how one aspect of autonomy for residents is their privacy and this is an important concern for many residents and families (16). To solve this problem, research groups have begun using depth-based cameras in order to maintain peoples privacy, such as Gasparrini, Cippitelli, Spinsante, & Gambi, 2014 (17). Depth-based cameras allow a person to show up on the camera without any identifiable features. It also identifies the different objects in the room, and can tell how far away they are from the camera. All of these features allow the camera to create an accurate picture of the
4


room and identify where the person is while maintaining their privacy. Gasparrini et al., (2014) used a Microsoft Kinect to use a frame-by-frame analysis of the room. The system extracts elements and classifies them as objects or as a person. Once the person was identified, a tracking algorithm was put into place in order to monitor their movements. The tracking algorithm allowed the system to handle the individuals interactions with stationary objects. When the person was detected near the floor, the system predicted they had fallen (17). Despite advances in privacy with this system, the researchers stated there are still two major problems that have not been solved using depth cameras. These include the number of cameras required to cover an entire room; the computational power required to sort through all of the data; and, the inability of camera systems to handle more than one person in the room. Each of these challenges present a large hurdle for camera systems to overcome in order to accurately detect falls and make this technology useful.
Smartphones
Other technologies under exploration are accelerometers, gyroscopes, and magnotometers built into todays smart phones. These sensors detect movements of the user and are implemented in many health apps that are found on phones today. Quite a few research groups have begun investigating this technology to determine its accuracy and usability and will be discussed in the following paragraphs (18, 19,20).
In a study completed by He, Li, & Bao, (2012), the group utilized the built-in triaccelerometer to collect information about body movement (19). They determined that body motion could be classified into five different patterns: vertical activity, lying, sitting or static standing, horizontal activity and fall. When this system detects a fall, it sends a
5


text message to a pre-selected group of people. The text message includes time, GPS coordinates, and a Google map of the suspected fall location. During testing for this system the phone was worn at the waist of the user. The researchers mentioned that it presented a problem because most people do not keep their phones mounted at the waist. Instead, many people keep their phones in their pants or shirt pocket. They also mentioned that in future studies they will need to solve the problem of adapting the technology to daily wear, and that it needs to work when the user places their phone somewhere besides their waist.
In a study completed by Madansing, Thrasher, Layne, & Lee, (2015), the smart phone system was divided into three phases: basic architecture, analysis, and communication. The basic architecture is the flow diagram of how the smartphone system works. The accelerometer, gyroscope, and barometer in the phone output data is collected, analyzed, and communicated via an email, text, or a phone call. The analysis portion described how the collected data from the phone is then analyzed and interpreted in order to determine if a person has fallen or not. The final phase is communication. In this step the system notifies the corresponding individual of the fall over email, text, or a phone call. This study went on to discuss the different locations that the smartphone could be placed for fall detection, such as the waist or chest. They also stated that the components found in a smartphone that could be used for fall detection are temperature gauges, touch sensors, cameras, and ambient light sensors. They then analyzed 25 different studies that used smartphones for fall detection. What they found was that each of these studies used only simulated falls, an android was the most accurate fall detecting system, and the accelerometer was the most used sensor followed by the gyroscope.
6


This study also found quite a few challenges with the smartphone technology itself. For example, the quality of sensors inside phones were not high enough to produce consistently accurate data. The energy consumption and battery life of this technology was very high and required a lot of time to recharge. The final challenge was placement and usability, because carrying a smartphone in a single pocket or position is difficult to do. The researchers mentioned that while this technology is incredibly convenient and easily accessed on a phone, it is not feasible for the elderly population nor is it practical for individuals with dementia. They discussed how the current aging population did not grow up with cell phones and most of them do not own a cell phone or know how to use one. Furthermore, they are not used to carrying it around every where and for elderly individuals who have memory loss there is a very good chance they will not remember where their phone is or to keep it in one position all day (18). This is the main reason why using the accelerometers, gyroscopes, and magnotometers inside phones is not practical for this population.
Microphones
Another type of technology that is an option for fall detection is sound sensors and microphones. These have become more and more popular because unlike cameras, they are able to maintain peoples privacy, and can be discretely installed in any room of a house or building. In a study completed by Li, Ho, Popescu, & Skubic, (2014) they installed 8 microphones on a circular wooden board and hung it on a wall (21). The system was comprised of three different processing steps: sound source localization, increase the signal to noise ratio, and fall recognition. This study consisted of 12 falls and 12 non-falls and found that analyzing sound at the beginning of the signal resulted in a
7


more accurate identification of a fall or non-fall. But a study by Li, Popescu, Ho, & Nabelek, 2011 (22) found that one limiting factor was that it operated under the assumption that there was only one person in the room. Additionally, it required a lot of microphones set up around the room in order to filter out the extra sounds, such as phone calls, typing, or a TV. Studies such as Li, Zeng, Popescu, & Ho, (2010) (23) found that it took an 8-microphone linear array to most accurately detect all of the extra reverberating sound. The same group also found that it takes at least a 2-microphone array in order to somewhat accurately locate sounds in a room (21). Based on the studies described above it would be difficult to use this technology at memory care facilities with many rooms, increased noise levels, and multiple people moving in a room at once.
Floor Sensors
One common sensor that is easy to use and has been considered in elderly communities is a pressure sensor. There have been a few research groups, which will be described in the following paragraphs, that have used these sensors by embedding them in carpets (24, 25, 26). They integrate multiple pressure sensors within a square carpet piece. A fall is detected on the carpet squares when a force is being applied to 2 or more pressure sensors in a single square.
One study completed by Aud, Abbott, Tyrer, Neelgund, Shriniwar, & Devarakonda, 2010 developed a smart carpet that consisted of an array of four sensors that are battery independent and placed in the carpet. To test the carpet they created a 10ft long prototype of the carpet. Eleven volunteers between the ages of 20 and 60 walked on the carpet and answered a few questions about it. None of the volunteers could identify a difference between the regular carpet and the smart carpet. The carpet did have a number
8


of false positives and false negatives regarding whether a person was stepping on the sensor or not during the trial. The carpet was able to determine a number of characteristics about a persons gait such as velocity and frequency, but no actual falls were tested on the carpet. The research team concluded that the smart carpet was able to detect human gait characteristics and that detection of a fall, which they would deem as a larger number of sensors triggered than when it was just a footstep, was feasible (26). Despite these conclusions, some concerns about this technology are that it has not been tasted with actual falls. It also has only been tested with one person in a room on one piece of carpet, and does not take into account a room with multiple wheelchair users.
In another study completed by Chaccour, Darazi, Hajjam, El Hassani, & Andres, 2015, piezo resistive pressure sensors were used in the carpets (25). The system was divided into two parts: the first was the pressure sensors that detected the signal, and the second was the processing of that signal. This system utilized a threshold-based algorithm to detect a persons falls, and then sent an SMS alarm notification in case of a fall event. In the experiment for this study they are examining falls and daily activities. For the experimental falls the volunteer was instructed to fall on his side and cover at least three different sensors. Elders were not considered in this experiment to avoid injury. The result was that the sensors triggered an alarm 8/9 times, and did not signal an alarm when the volunteers ran, walked, jumped, or sat on the carpet (25). This was a very limited experimental study due to the fact that the volunteers had to position themselves in a specific way in order to ensure the alarm went off.
In these studies, the sensors have been somewhat accurate in determining if someone has fallen or not, but none of have been used in a real world community or
9


tested what will happen when there is more than one person walking on the carpet. Another problem is that many dementia facilities prefer hard wood floors, especially in bedrooms and bathrooms where falls are most likely, because it is easier to clean up common incontinence accidents (27). So while this technology has a lot of benefits and potential, it is not entirely appropriate for the population of people with dementia. Doppler Radar
A method that has become very popular in the past few years for detecting falls is using Doppler radar because it maintains an individuals privacy, is accurate, and can be discretely placed around the room so people do not know it is there. Originally, this technique did not start out very precise, Uegami, Iwamoto, & Matsumoto, 2012 found it was around 88-96% accurate (28). This study used a system that acquired and digitized the sensor output, processed the sensor data, and then transmitted the results to their computer. The movements that were tested were tripping and falling, walking, shaking arm and hand, using a cell phone, sitting on a chair, and, standing without moving. Each movement was tested 20 different times. The system resulted in 20 activities that were falsely identified as falls.
Liu, Popescu, Skubic, Rantz, Yardibi, & Cuddiby, 2011 study improved these results with a proposed fall detection technique that utilized two pulse-Doppler range control radars (29). They estimated the velocities of subjects within the detection range and tried to recognize a fall based on its Doppler signature and velocity. They used a pilot dataset of 109 falls and 341 non-fall human activities, and obtained an accuracy of 91% and 97% for detecting falls and non-falls respectively. But as the technology continued to develop Liu, Popescu, Skubic, & Rantz, 2014 (30) completed another study that found
10


the best location for these sensors was on the ceiling, and that it was even more accurate when combined with a motion sensor.
Additionally, Garripoli, Mercuri, Karsmakers, Soh, Crupi, Vandenbosch, & Schreurs, 2015 (31) was able to use this technology to detect 100% of falls with complete accuracy. In this study, the system that was tested consisted of a sensor, combining radar, computational, wireless communication, and a base station for data processing. The experiment was performed in a 5m x 5m room with one volunteer at a time. There was furniture in the room to mimic a real world room setting. The sensor was fixed to a wall at a height of 1.25m while the base station was positioned 4m away from the sensor. The system was validated by having 16 volunteers, 14 males and 2 females, simulate 65 fall events. Each volunteer was monitored for 5 minutes total including before and after the fall. Additionally, 40 random walking activities, 30 activities of sitting down or standing up, and 20 random movements such as closing windows and moving a chair were also observed. The results were that there were no false positives, but they did find that the farther away the subject was from the Doppler radar the more likely they were to get lost in the noise of the objects around them. While this was a significant advancement, the biggest remaining problem with Doppler radar is that it cannot detect if a person has fallen when the person is obstructed by objects such as a bed or couch Accelerometer/Barometer/Gyroscope Sensors
The most researched type of technology for fall detection are accelerometers, gyroscopes, barometers, and magnetometers (32). Each of these sensors have been used either separately or together in order to collect data about the users body position, detect if they have fallen or not, and then send an alert that the user has fallen (33,34,35, 36).
11


One of the biggest challenges for using these types of sensors in fall detection is determining the best location to place them on the user. Most researchers have determined that the most accurate location is the waist, trunk or head due to less movement in these areas, compared to feet, legs, or wrists (37).
Early research used a tri-accelerometer to detect individual body motion in the X, Y, and Z-axes. Bourke et al., 2010 (33) performed a study using a single accelerometer at the waist in conjunction with a threshold algorithm. This threshold algorithm used specific thresholds set for the kinematic and angular data collected, and when a specific combination of thresholds was reached, it would signal a fall had occurred. In this study, multiple threshold algorithms were created in order to determine which one was the most accurate. The device was placed on 10 young healthy subjects who performed 240 falls from various starting points and directions and 120 activities of daily living. The device was also placed on 10 elderly healthy subjects who performed 240 scripted and 52.4 hours of continuous unscripted normal activities. The results showed that the algorithm combining the parameters of impact, posture, and velocity achieved the lowest falsepositive rate of .94 false positive per day with a sensitivity of 94.6% and a specificity of 100%. Sensitivity was defined as the percentage of falls that were correctly identified as falls, and specificity as the number of daily activities that were accurately classified. Overall, this study was successful and the use of elderly individuals was very valuable, but it may not be practical for a large community of patients with dementia. The false positive rate is very high for a larger population, and this device was worn at the users waist making it obtrusive for this sensitive population.
12


Another group that used a similar threshold-based fall detection algorithm was Dumitrache & Pasca, 2013 (38). This group used the algorithm to compare the thresholds of six different parameters during 34 simulated falls and 200 daily activities that resulted in a sensitivity of 97.05% and a specificity of 99%. Overall, the algorithm of this study was extremely accurate. The major limitation of this study was that one person was used to complete all of the simulated falls and daily activities. Furthermore, the individual was a young man, so it was not an accurate representation of the elderly population or the population with dementia. Another problem, as it relates to a population with dementia, is that it requires the user to remember to wear the device and to place it at their waist daily.
A study performed by Chen, Feng, Zhang, Li, & Wang, 2011 (34) used a single accelerometer and a new algorithm to develop a device that had more accurate sensitivity with more movements. In this study, the researchers performed 195 simulated falls and 75 daily activities that resulted in a sensitivity of 97% and a specificity of 100%. The accelerometer was placed in a large square box that was worn on a belt at the users waist. The algorithm, which was more accurate than previous studies, measured the total sum acceleration and the tilt angle of the body during different phases of a fall event.
This algorithm included three thresholds and three delay times, which helped determine if the body motion and posture indicated a fall or not. Despite success with this algorithm, this study did not use its device on any elderly individuals, which may alter the number of falls that were detected. It also still required the individual to remember to wear a large obtrusive device at their waist.
Lan, Hsueh, & Hu, 2012 (39) performed a study with one accelerometer and a different type of algorithm than previous studies. This study used a support vector
13


machine framework to process the accelerometer data. This method could only discern up to 4 different kinds of fall events and failed to recognize 5 falls out of 120, which resulted in an accuracy of 95.83%. This system was less accurate and less robust compared to the earlier studies. It also required the device to be worn in a box on the waist and was only tested on young subjects.
The next evolution of this technology combined accelerometers with gyroscopes and barometers to help improve the detection of body movement and location. Tolkiehn, Atallah, Lo, & Yang, 2011 (40) performed a comparative study using an accelerometer alone to using an accelerometer combined with a barometer. In this study, measurements included features of the body tilt while accelerating, the impact magnitude, and the body tilt angle while no acceleration change occurred. When a fall was detected using these features, it was verified with the barometric pressure sensor data. It determined if the pressure rose at the same time the suspected fall occurred; validating if the person actually fell. The study included 12 young healthy subjects, and each had to perform 13 different fall activities and 11 daily living activities. Following each simulated fall onto a mattress, the subjects were told to remain fallen for 15-25 seconds to simulate the lying down period that some elderly people experience after falling. The accelerometer alone was 81.48% accurate, 83.33% specific, and 79.08% sensitive, whereas the accelerometer and pressure sensor combined resulted in 86.97% accurate, 85.24% specific, and 87.77% sensitive. The results from this study showed that using an accelerometer and a pressure sensor was more accurate in detecting falls than using an accelerometer alone.
In a study by Bianchi, Redmonds, Narayanan, Cerutti, & Lovell, 2010 (41), a barometer was combined with an accelerometer to create a sensor that was worn at the
14


hip to detect a fall. This study performed three different tests to determine if the sensor accurately detected falls. The first test was comprised of only indoor movements and falls, the second test only outdoor simulated falls, and the third test indoor and outdoor simulated normal activities. The study used 16 different movements, 8 different fall movements, and 8 different daily activities. In these experiments, all simulated falls landed onto a mattress. This study included 20 young healthy volunteers. Three different types of algorithms were tested to see which one had the highest accuracy of fall detection. The first algorithm was based on the extreme impact of a fall and movement intensity that can be obtained from calculating the signal magnitude vector (SVM). When the SVM reaches a signal higher than 1.8 G, it was considered a fall. The second algorithm was based on the first algorithm, but assumed that a fall always ends with a non-standing body orientation of the wearer. Using this system, a fall is detected if the data indicates a large impact followed by a non-standing body position. The third algorithm takes into consideration the barometric pressure measurement when determining if a fall has occurred. It follows the same assumptions as the earlier algorithms, but adds the change in altitude of the device at waist level when a fall occurs. This study resulted in an accuracy, sensitivity, and specificity of 96.9%, 97.5%, and 96.5% respectively using algorithm three. This was an improvement from studies using only an accelerometer. However, fall detection devices could be improved by testing with an elderly population, continued improvement of accuracy, making the device smaller when worn at the hip, and performing a larger number of falls and daily activities to validate accuracy (41).
15


Another study that used both an accelerometer and a barometer was Wang, Narayanan, Lord, Redmond, & Lovell, 2014 (42), whose main goal was to create a system that not only detected falls, but also used less power. This study included 20 young individuals who performed 8 simulated falls and 8 daily activities. The individuals wore the device at their waist, which detected falls using thresholds from both the accelerometer and the barometer. The device was low power, sampling at 5Hz and then increased to 40 Hz when the data met the threshold criteria to signal a fall. Overall, the results of the study were promising in that it showed 95.9% accuracy, 96.7% sensitivity, and 96.9% specificity. It also showed that the original algorithm, without changing sampling frequency, was about 1% more accurate. This study did not use elderly individuals in their sample and was still worn obtrusively at the hip. However, the sacrifice of accuracy for power was an important concept to consider since this device was worn daily.
The most recent advancement in using accelerometers, gyroscopes, and barometers for fall detection has been the combination of the three sensors into an inertial measurement unit (IMU). These units often contain a three-axis accelerometer, three-axis gyroscope, and a three-axis magnetometer, but some units contain additional advanced features. The IMU is an important advancement, because it makes the overall system much smaller and less obtrusive. This combination sensor has recently become more popular and frequently utilized in fall detection research. Pierleoni, Belli, Palma, Pellegrini, Pernini, & Valenti, 2015 (43), conducted a study that used this sensor and tested it against multiple algorithms. In this study, an IMU sensor that included a tri-axial accelerometer, gyroscope and magnetometer with a microcontroller, Bluetooth module,
16


mass storage unit, battery, and wireless receiver in the form of a phone or computer was used. Different algorithms were used from two other previous studies that only looked at impact and posture. An additional algorithm was created that looked at the impact, aftermath, and posture. The new algorithm extracted the acceleration and pitch and roll angles, which are rotations around the Y and X-axes respectively, and compared them to threshold values. These were selected after an accurate training process using the support vector machine method was completed. This training method consisted of 50 simulated falls and 50 daily activities. The sensor was placed in a small hard box that was worn on the waist, and tested on 10 volunteer subjects from the ages of 22 to 29. Each subject repeated 18 scenarios comprised of nine simulated falls and nine daily activities replicated three times, resulting in 540 tests. Subjects performed a simulated fall onto a mattress to ensure they were safe, and no elderly subjects were used in this test. The new algorithm performed better than all other algorithms it was compared with. It had 100% accuracy, 100% sensitivity, and 100% specificity when using one experimental protocol, and 90.37% accuracy, 80.74% sensitivity, 100% specificity when using a second experimental protocol. The difference in the two protocols was that the second protocol included 2 falls where the subject ended up sitting with their back staying vertical to the ground. The falls in the first protocol all ended with the subject lying down with their back horizontal to the ground. Additionally, the algorithm took into consideration the position at the end of the fall, which indicated that this algorithm was able to more accurately detect the typology of the fall and the end position of the user. Overall, this study proposed a very accurate algorithm that was able to detect most, if not all, falls and provided a good indication of the end position of the fall. The limitations to this study
17


were that this method was not tested on an elderly population and it had to be worn at the waist. The addition of a barometer to this system may further improve the location and position accuracy due to its ability to determine a change in height.
Impact of Technology to Detect or Predict Falls Fall detection systems have made significant progress, but few have been specifically designed for individuals with dementia or for use in a memory care facility (36). Those with dementia are very sensitive to sensory stimuli and regularly deal with memory loss, decreased eyesight, pain in joints and legs, and a decreased ability to communicate or understand others, which can result in tremendous frustration (44). A fall detection system needs to be developed that takes these health conditions into consideration so that it can be used successfully with this population. Because of the complexity of caring for individuals with dementia, most families decide that the best place for their loved one is in a memory care facility with caretakers who have extensive experience (45). By 2050, the number of people using long-term care services will likely double from 13 million in 2000 to 27 million people (45). This is mostly due to the increasing number aging people (45). One of the biggest challenges facing these facilities will be keeping track of the number of falls that occur (45). Typically, long-term care facilities follow a protocol called Standard of Care. These are a set list of guidelines that the facility must follow in order to properly care for and monitor the residents, especially if they are at an increased risk of falling. These guidelines are defined by broad requirements set by federal and state law under 42 CFR §483 of the Nursing Reform Act. Under this protocol, the long term care facility checks on the residents once every hour to ensure that the resident is okay and has not fallen. Despite this procedure there are still
18


many falls that go unseen and unreported (45). Due to this there is a critical need to design a real-time, accurate, wireless fall detection device to detect falls that occur from a standing position by a resident with dementia in a memory care facility.
Partnership with Anthem Memory Care In collaboration between Dr. Bodines Bioengineering Labs at University of Colorado Denver and Anthem Memory Care facilities, the research team identified a priority to address currently undetected falls of residents with dementia living at the facility. Despite efforts with current standard of care protocols, there are still falls that go unseen and may or may not be reported after the fact by the resident or by other residents who may have witnessed the fall. The question that this study proposes is: Can a realtime, wireless fall detection device detect falls from a standing position and be worn for 3 days by a resident with dementia in a memory care facility? Additionally, the goal is that this device will be used in addition to existing standard of care protocols that Anthem Memory Care has in place to improve quality of care for residents with dementia.
19


CHAPTER II
SPECIFIC AIMS
Specific Aim One. Design and build a real-time wireless fall detection device using an IMU sensor, microcontroller, and battery.
Specific Aim Two. Perform an iterative test using the device on young volunteers between the ages of 20-40. These volunteers will mimic eight different falls and complete five daily living activities, which include lying on a bed and then standing, walking a few meters, sitting on a chair then standing, climbing two steps, and standing after picking something up. These tests will ensure there are no false alarms when differentiating between falls and daily activities.
Specific Aim Three. Build and test a mock wireless fall detection device modeled for the size and weight of the fall detection system (FDS). The purpose is to test the mock device on 20 residents at Chelsea Place Anthem Memory Care to demonstrate that the FDS design is unobtrusive to activities of daily life and to ensure that the actual device will not be damaged in the next phase of testing.
Specific Aim Four. Test the FDS by integrating the device into a 3 in x 4 inch Band-Aid that will be worn by 20 residents at an Anthem memory care facility. The device will operate in addition to the current standard of care. The purpose is to test the FDS with the intended population to validate that the device can detect falls that occur from a standing position by a resident with dementia at Chelsea Place Anthem Memory Care.
20


CHAPTER III
MATERIALS AND METHODS Fall Detection System Development
The objective of the fall detection system development was to acquire, a microcontroller, 10-Degree of Freedom Inertial Measurement Unit with a L3GD20H + LSM303 + BMP 180 sensor, and battery to be assembled into an initial prototype. A 10-degree of freedom inertial measurement unit is a sensor that has a gyroscope (L3GD20H), accelerometer/magnetometer combo (LSM303), and a barometric pressure sensor (BMP 180). Then, a fall detection system was developed. This required the development of an algorithm that could collect acceleration and change of altitude from the users body movements and, using a series of four threshold values, determine if the user had fallen. After this is determined, the algorithm then sends the data samples, which include the values from the fall, to a secure broker called a Cloud MQTT. A MQTT (Message Queuing Telemetry Transport) is a messaging protocol that provides network clients with a way to distribute and communicate information from machine-to-machine using a publish/subscribe method. The Cloud MQTT then sends the data to a computer. The algorithm also connects to the Adafruit MQTT, which sends a notification that the user has fallen to a smartphone. The flow of this system is shown in Figure 1.
21


Figure 1. Block diagram of the fall detection system.
Protocol and Methods
The Adafruit Feather HUZZAH ESP8266 WiFi microcontroller was selected because of its small size and ability to connect to the WiFi. The Adafruit 10-Degree of Freedom Inertial Measurement Unit Breakout L3GD20H + LSM303 + BMP180 inertial measurement unit was chosen because it contained an accelerometer, gyroscope, and barometer. These sensors provided 10 degrees of freedom and five data points, yaw, pitch, roll, magnetometer, and barometric readings, to collect for fall detection. For the prototype these two pieces were wired together on a breadboard. The 3 V, Ground, SCL, SDA pins on the Huzzah Feather ESP8266 were connected to the 3V, Ground, SCL,
SDA pins on the Adafruit 10-DOF IMU Breakout respectively, as shown in Figure 2.
22


After the IMU sensor and microcontroller were connected in this manner, the IMU sensor
could send the microcontroller the onboard sensor data.
Figure 2. Adafruit Huzzah Feather ESP8266 wired to the Adafruit 10-DOF IMU.
A 1200 mAh 3.7 V Lithium Polymer battery powered the device because it is lightweight and has a long battery life. The Huzzah Feather ESP8266 has a built in battery regulator that only allows the battery to power the device until it reaches 3.14 V. Once it reaches this voltage the Huzzah Feather ESP8266 automatically shuts off. This results in a battery voltage range of 4.2 V to 3.14 V. Tests to determine battery life were performed using a multimeter to confirm this range. Using a voltage divider circuit, seen in Figure 3, the battery voltage was stepped down to a max of 0.9 V and a minimum of 0.6 V. The output voltage from the voltage divider was connected to the ADC pin on the Adafruit Feather HUZZAH ESP8266. The voltage was stepped down because the ADC
23


pin can only read analog voltages between 0 V and 0.9 V. This pin converted the voltage into a digital range of 680 to 780. Two tests were completed to help determine how battery life is affected by frequency of data deliver. The first test recorded the output value of the battery every 0.5 s into a 1000 row matrix. This matrix was outputted into an excel spreadsheet every 8 min until the battery died. The second test recorded the output value of the battery every 1 s, and sent the value directly to an excel spreadsheet immediately after the data point was recorded. After the battery life tests were finished, a fall detection algorithm was developed using C++ code on the Arduino platform.
Figure 3. (a) Simulation of voltage divider circuit, (b) Voltage regulator circuit connected to Fall Detection Device Prototype.
The algorithm uses the accelerometer readings, yaw, pitch, and roll, which are identified as rotations around the Z, Y, and X-axes respectively. These measures determine the orientation of the body, and can be seen in Figure 4. It takes a sample of this output data every 0.25 s. It stores up to 400 rows of data samples in a matrix. When this limit is met it starts re-placing the data in the matrix starting with the first row.
24


y
Figure 4. Reference system for the Yaw, Pitch and Roll angles (43).
The algorithm examines each sample using a threshold technique in order to look at the impact, aftermath, and posture phases of a fall. The thresholds and definitions for these measures were based on Karatonis studies (46), which were also verified by Pierleoni et al., 2015. These values can be seen in Table 1 below. The definitions of these thresholds are as follows:
Impact Phase: is the time after the loss of foot contact with the ground and the start of falling towards the ground, due to attraction of gravity force; the subject impacts on the ground, or other objects that cause an acceleration peak
Aftermath Phase: is the immobilization, or nearly so, to the ground associated with low values of acceleration for a short time
Posture Phase: is the end posture, dependent on the direction of the subjects trunk after falling to the ground
The algorithm looks at the impact threshold first, then the aftermath threshold, and then finally the posture phase threshold. An overview of the algorithm is shown in Figure 5.
25


Figure 5. Flow chart of the proposed fall detection algorithm.
The impact phase threshold is set at 4 g. At any point when the root mean square of the acceleration is above that threshold it indicates that the impact phase of a fall may have occurred. The root mean square is examined for this threshold because the single combined acceleration value gives a more accurate representation of how the whole body is moving, which increases the probability of detecting falls. If this threshold is met the algorithm completes 9 more iterations, at a sampling rate of 0.01 s, of data collection. Each of these data samples is examined to determine if they meet the next threshold. If they do not, the code resumes sampling data every 0.25 s and looking for a spike in the root mean square value.
26


The next threshold that is examined is the aftermath phase. It is detected when the root mean square of acceleration presents an almost flat trend close to 1 g. The algorithm checks the lower and upper bounds, 1 g and 2 g respectively, of the threshold for 1 s following the impact threshold being reached. It also checks to see if there was an altitude change of 0.4 m. If this threshold was met, the algorithm moves onto the last criteria. If it was not met, the code finishes sampling and examining the rest of the 9 data points.
Once finished, it then moves back to sampling data every 0.25 s and looking for a spike in the root mean square value.
The last threshold is the posture phase, which is defined by orientation, which is an
angle within a specified range within 1 s after the aftermath phase. The posture phase is
observed by looking for changes in pitch and roll. The pitch and roll values change based
on whether the person has fallen right, left, backwards, or forwards. The specific values
for each direction are listed in Table 1. If this final threshold is met, the code assumes
that a fall has occurred. At this point, the algorithm then connects to WiFi and connects to
the Cloud MQTT and Adafruit MQTT. An additional 100 data points are then recorded,
at a sampling rate of 0.001 s, to ensure that the data from the entire fall was recorded.
Table 1. Shows the phases of a fall, the important parameters of that phase, and the corresponding threshold values that are being used.
Phase of Fall Parameter of Interest Threshold Value
Impact Phase Mean Acceleration x>4 g
Aftermath Phase Mean Acceleration 1 g Posture Phase Pitch and Roll Fall Left: -70 27


It then takes the 400 data points that are populating the matrix and packages it into a JSON packet. JSON (JavaScript Object Notation) is a standard format to send data objects in an attribute-value pair. One row of the JSON packet contains: RMS: Corresponding Value, Altitude: Corresponding Value, Roll: Corresponding Value, Pitch: Corresponding Value, and Time: Corresponding Value. This is repeated for each of the 400 rows. This packet is then encrypted and securely published to the Cloud MQTT under the name Fall 2. It was encrypted and securely published using a Secure Sockets Layer port between the Cloud MQTT and its connected clients. It is then sent to any client that is connected to Cloud MQTT and subscribed to accept any message or data with the name Fall 2. In this case, the client is a computer that is connected to Cloud MQTT using a python program, and is subscribed to receive any messages that are sent to the Cloud MQTT server with the title Fall 2. The Cloud MQTT takes the message from the fall detection sensor and then sends it to the computer. This entire process takes about 0.18 s. The python code then takes the JSON packet and examines it one row at a time. It separates the values in the row based on the labels. It places the separated pieces of data into the corresponding heading within the comma-separated values (CSV) file titled test.csv. The fall detection device also sends the number 1 to the Adafruit MQTT feed fall3. This changes the feed fall3 from 0 to 1. This triggers the application IFTT, which was monitoring the feed fall3, to send the notification Fall has occurred to the smartphone, as seen in Figure 6. IFTT stands for If This Then That and is a web-based service that allows users to make conditional statements called Applets. The Applets work by allowing the developer to dictate if this happens then that happens. In this case,
28


if Adafruit feed fall3 equals 1 then a notification message Fall has occurred is sent to the smartphone with the IFTT application. It can take 2 to 5 minutes for this notification
to occur.
$ 83%
Q. Search #
Recent -
^ IFTTT now |
Fall Detected on Sensor 1 k _
Figure 6. Fall detection notification shown on a smartphone.
Figure 7 shows how the Fall Detection Device, Computer, and IFTT Application work together with the Cloud MQTT and Adafruit MQTT.
29


Figure 7. Block diagram showing how the client/broker system works.
Young Healthy Volunteer Study
The objective of the young healthy volunteer study was to determine if the fall detection system could accurately and consistently detect 9 different types of falls and not classify four different activities of daily living as falls.
Sites
The assistive technology partners lab was used to test the fall detection system on the young healthy volunteers. The set up for the study, which can be seen below, consisted of two teal pillows, and two couch cushions covered by the yellow and blue blanket. These are on top of two gym mats commonly used in gymnastics.
30


Figure 8. Set up of the test site at assistive technology partners lab.
Subjects
The study was completed using 18 healthy individuals between the ages of 20-56 who have worked with people with disability. Table 2 shows age, gender, and height Table 2. Demographic information about the subjects.
Gender Height Age
Female 57 23
Female 5'3 24
Female 5'0 24
Female 5'8 23
Female 5'6 25
Male 5'9 25
Female 5'4 24
Female 5'9 23
Male 67 24
Female 5'4 23
Male 5'6 23
Male 6'4 26
Male 67 26
Female 5'5 55
Female 57 56
Female 5'9 28
Female 570 30
Male 6'0 55
31


Consenting
Study included people testing the prototype by completing falls and being video taped with a camera. Subjects who were video taped at ATP consented by signing the Colorado Multiple Institutional Review Board approved consent forms under the study, An Exploratory Investigation of the Impact of the Assistive Technology Partners Product Testing Lab. These forms included a consent packet and a photograph/videotape release form. The form were read by each subject and reviewed verbally in a quiet section of ATP before signing.
Protocol and Methods
Following the Usability Protocol (COMIRB #11-067), the study was completed using 18 individuals between the ages of 20-56 who have worked with people with disability. Participants wore a strap around their chest and the FDS was securely placed underneath the strap for the duration of the study, which is shown in Figure 9 below.
w
Figure 9. The blue square shows how the fall detection device was placed under the strap that was secured around the participants chest.
32


The participants completed 9 different fall types into a soft padded mattress. The types of falls are listed below:
Fall backwards and end on left or right shoulder
Fall backwards and end on back
Fall forward and end on stomach
Lateral left fall and end on stomach
Lateral left fall
Lateral right fall and end on stomach
Lateral right fall
Fall backwards, land on butt, and end lying on back
They were then asked to complete four activities of daily living, which are listed below:
Walk 5 meters
Sit down and stand back up
Bend down to pick up a pencil from the floor and stand back up
Lay down and then stand back up Data Collection
The components in the inertial measurement unit recorded the subjects body movements. The accelerometer recorded the subjects acceleration in the X, Y, and Z-axes. The gyroscope and magnetometer readings were recorded to help determine if they were sitting, standing upright, or lying down, and the barometric readings were recorded to determine if they are still standing or not. When each of these components reached a specific threshold a fall was detected. Once a fall was detected 400 data points, 180 before the fall and 20 after the fall, were sent to a computer excel spreadsheet.
33


Data Analysis
Once each subject had completed the protocol, the number of falls detected by the device, the number of falls that were not detected but were witnessed by the principle investigator, and the number of false alarms the device detected were tallied. False alarms are defined as the device sending a fall alert when the user did not fall. Then, sensitivity, specificity and accuracy were calculated as described below:
1) Sensitivity: Capacity to detect a fall, which is calculated as the ratio of true positives and the sum of true positives and false negatives
2) Specificity: Ability to avoid detection of a normal event as a fall, which is calculated as the ratio of true negatives and the sum of true negatives and false positives
3) Accuracy: Capacity of correctly detecting a fall, which is calculated as the ratio of true positives and the sum of true positives and false positives
Additionally, the accuracy, average, and standard deviation of the fall direction data were calculated. Visual graphs of each fall direction were also created to show the difference between each direction. The falls that were not detected were examined in order to determine why that was the case.
Mock Device Study
The mock device phase of the study was completed to determine whether the device could be integrated into the daily routine of residents who live in a memory care facility without causing irritation. It also tested whether or not the FDS, which is more costly, would be destroyed when it was worn for testing.
34


Sites
The sites used for this study were Chelsea Place Anthem Memory Care in Aurora. Subjects
10 residents at Chelsea Place Anthem Memory Care were recruited to wear the mock device in addition to continuing to receive standard of care. 9 of them agreed to participate and their demographics are listed in Table 3. Inclusion and exclusion criteria for this population are listed below.
Inclusion
Diagnosed with dementia
Willingness and capability to give informed consent to participate in the study, or consent of a legally appointed guardian if he/she is unable to provide full consent according to institutional guidelines
Primary language is English
Living at Chelsea Place Anthem Memory Care
Is able to walk around Chelsea Place Anthem Memory Care on their own Exclusion
Residents at Chelsea Place Anthem Memory Care facility that primarily use a wheelchair
Residents at Chelsea Place Anthem Memory Care facility that are on end of life hospice care
Residents at Chelsea Place Anthem Memory Care facility that have a history of developing skin rashes from Band-Aids
35


Residents at Chelsea Place Anthem Memory care who have a current injury that prevents mobility
Table 3, Resident demographics; gender and ambulatory status.
Subject # Sex Age Diagnosis Current Ambulatory Status
1 Male 84 Alzheimer's Walks with walker if he remembers
2 Male 79 Alzheimer's Walks on his own
3 Female 70 Alzheimer's Walks on her own
4 Female 95 Alzheimer's Ca ne/wa 1 ke r/wa 1 ks on her own
5 Female 90 Alzheimer's Walker
6 Male 88 Alzheimer's Cane
7 Female 81 Vascular Dementia Walks on her own
8 Female 87 Alzheimer's Walks on her own
9 Female 75 Alzheimer's Walks on her own
Consenting
After the legally authorized representative (LAR) volunteered, the LAR was asked to meet with the trained study staff/investigator to read, ask questions, and sign the consent form. This meeting occurred in a quiet, private meeting room at Chelsea Place Memory Care facility. The room was outfitted with a table, chairs, and any necessary accommodations needed by the potential subject to comfortably and fully participate in the consenting process (e.g. large-print documents for someone with a visual impairment) with no external distractions.
The trained study staff/investigator read through the consent form with the LAR. The LAR was asked if they had any questions or concerns, and was informed and assured that
36


they could remove their loved one from the study at any time and no one would be mad. Once their questions had been answered, the trained study staff/investigator left the room and allowed the LAR to think about whether they wanted their loved one to participate in the study. When a decision was reached, the LAR opened the door to the room to signify that they had reached a decision, and the trained study staff/investigator came back into the room to hear the decision. If the LAR decided to participate, all required parties signed the consent form.
Protocol and Methods
9 subjects wore the mock device for 3 days. The mock device was the same size and weight as the actual fall detection system. The total weight was approximately 35 grams. It was integrated into a Band-Aid intended to avoid irritating the skin, and was similar to existing patches that some residents wear for pain relief. It was worn by residents on their mid-back or lower shoulder (see Figure 10). The trained carestaff at Chelsea Place Memory Care placed the Band-Aid and monitored them daily. If the resident became irritated or agitated by the device, or developed a skin irritation from the device, the carestaff were instructed to immediately remove it.
Figure 10. Mock device that was placed on the mid-back or lower shoulder of subjects.
37


Data Collection
The carestaff monitored the mock device on the resident throughout the day. If a user was irritated or developed a skin irritation from the mock device it was reported to the PI and they were removed from the study. If the mock device fell off overnight, while walking around, during an organized activity, or while sitting it was reported to the PI.
Final Device Testing
The final device testing ensured that the system could be integrated into the daily routine of residents at a memory care facility without it being aggravating or irritating, while also detecting any falls that occur while they wore the device.
Sites
The sites used for this study were Chelsea Place Anthem Memory Care in Aurora. Subjects
These subjects are the same subjects as the mock device study, excluding anyone that became irritated by device or developed a skin rash from it. Inclusion and exclusion criteria for this population are listed below.
Inclusion
Diagnosed with dementia
Willingness and capability to give informed consent to participate in the study, or consent of a legally appointed guardian if he/she is unable to provide full consent according to institutional guidelines
Primary language is English
Living at Chelsea Place Anthem Memory Care
38


Is able to walk around Chelsea Place Anthem Memory Care on their own
Exclusion
Residents at Chelsea Place Anthem Memory Care facility that primarily use a wheelchair
Residents at Chelsea Place Anthem Memory Care facility that are on end of life hospice care
Residents at Chelsea Place Anthem Memory Care facility that have a history of developing skin rashes from Band-Aids
Residents at Chelsea Place Anthem Memory care who have a current injury that prevents mobility
Table 4, Resident demographics; gender and ambulatory status.
Subject # Sex Current Ambulatory Status
1 Male Walks with walker if he remembers
2 Male Walks on his own
3 Female Walks on her own
4 Female Cane/walker/walks on her own
5 Female Walker
6 Male Cane
7 Female Walks on her own
8 Female Walks on her own
9 Female Walks on her own
Additionally, all carestaff that worked with the subjects in this study were asked to complete a consent form and take a short usability survey.
39


Consenting
The consenting process completed in the mock device study carried over to the final device testing. The caretakers that volunteered to take the survey were consented after the final device testing was completed. The consent meeting for any caretakers who volunteered to take the survey took place in a quite private meeting room at Chelsea Place Memory Care facility. The trained study staff/investigator read through the postcard consent form with the caretaker. The caretaker was asked if they have any questions or concerns, and was informed and assured that they could choose not to fill out or finish the survey at any time and no one would be mad. Once the questions had been answered, the trained study staff/investigator left the room and allowed the caretakers to think about if they are willing to participate. When a decision had been reached, the participant opened the door to the room to signify they had reached a decision and the trained study staff/investigator came back into the room to hear the decision. If the caretaker decided to participate, they filled out the survey following the postcard consent form.
Protocol and Methods
The final phase of this protocol included having 9 residents wear the FDS, seen in Figure 11, for 3 days. The total weight of the FDS was approximately 35 g. It was integrated into a Band-Aid intended to avoid irritating the skin and was similar to existing patches that some residents wear for pain relief. It was worn by the residents on their mid-back or lower shoulder. The trained staff at Chelsea Place Memory Care placed the Band-Aid and monitored it daily. If the resident became irritated or agitated by the device or developed any skin irritation, the carestaff immediately removed the device.
40


Figure 1
. Final Fall Detection Device that was placed on the mid-back or lower should
of the subjects.
Data Collection
The components in the inertial measurement unit recorded the subjects body movements. The accelerometer recorded the subjects acceleration in the X, Y, and Z-axes. The gyroscope and magnetometer readings were recorded to help determine if they were sitting, standing upright, or lying down, and the barometric readings were recorded to determine if they are still standing or not. When each of these components reached a specific threshold, a fall was detected. The fall data was then sent to a computer, an alert that the resident had fallen was triggered, and a notification was sent to the caretakers Anthem Memory Care standard issued iPod. Facility staff then implemented their Standard of Care Fall routine, which is outlined below:
First your ward will be evaluated to determine if severe pain exists and if he/she can move. If he/she has severe pain they should not be moved, and the emergency services will be summoned. If an Anthem Memory Care nurse is in the community, an evaluation and assessment may be
41


completed prior to notifying EMS. If after the licensed nursing assessment is complete, the nurse may direct to NOT notify EMS. The Primary care physician must be notified.
Next, the Anthem Memory Care staff will determine if your ward is weight bearing and can assist staff members in getting them up from the floor.
Then if they feel like they can bear weight, two employees can provide standby assistance to help the resident get up from the floor. If possible, staff members should provide reassurance and calmly talk to them throughout the process.
If they cannot assist and is unable to bear weight, the care staff will make them as comfortable as possible and call emergency services. They will offer pillows for support, and blankets for comfort.
The care staff will stay with your ward at all times until emergency services arrive.
Next an Incident Report form shall be fully completed and placed in the Incident Report binder.
Repeated falls by your ward will have follow-ups to include a Fall Risk Assessment and progressive measures to help avoid future falls.
If your ward falls and strikes their head, there is a risk for a closed-head injury and, therefore, 911 is to be called and evaluate the need to be sent to the ER or seen by a MD for evaluation of the injury.
42


A Head Injury Monitoring Form is to be implemented for all falls with suspected or actual head injury involvement.
The daily reports about how the subject tolerated the device, and how well it stayed on the subject were also recorded. Additionally, the system usability surveys that the caretakers filled out were collected.
Data Analysis
Once each subject had completed the protocol, the number of falls detected by the device, the number of falls that were not detected but were witnessed by care staff in the facility, and the number of false alarms the device detected were tallied. False alarms are defined as the device sending a fall alert to the caretaker when the user did not fall. Then, sensitivity, specificity and accuracy was calculated as described below:
4) Sensitivity: Capacity to detect a fall, which is calculated as the ratio of true positives and the sum of true positives and false negatives
5) Specificity: Ability to avoid detection of a normal event as a fall, which is calculated as the ratio of true negatives and the sum of true negatives and false positives
6) Accuracy: Capacity of correctly detecting a fall, which is calculated as the ratio of true positives and the sum of true positives and false positives
Once this was completed, the scores from the system usability survey were tallied in order to determine what the caretakers thought of the device.
43


CHAPTER IV
RESULTS
Fall Detection System Development
The battery life of the 3.7 V 1200 mAh lithium polymer battery while running the fall detection algorithm without sending any data points was 36 hrs. The battery life while sending 1000 data points once every 8 min was 20 hrs. The discharge of the battery, seen in the figure below, was what is expected of a lithium polymer battery.
Time (hrs)
Figure 12. Battery voltage reading throughout the 20 hrs of battery life.
The battery life while sending the data point immediately after recording the point was 12 hrs. The accelerated discharge of the battery can be seen in the figure below.
44


Time (hrs)
Figure 13. Battery voltage reading during the 12 hrs of battery life.
It was also found that it takes .18 seconds to send the data from the fall detection device to the computer. Additionally, it can take 2 to 5 minutes for the fall detection device to send a notification to the smartphones through the IFTT phone application.
Young Healthy Volunteer Study
The 18 subjects completed 9 different falls and 4 activities of daily living (ADL). In total, 164 falls and 72 ADLs were completed. The fall detection system did not identify any of the ADLs as a fall. The number of true positives was 155, where a true positive is defined as a fall that was correctly detected. The number of true negatives was 72, where a true negative is an activity of daily living that is not misidentified as a fall. There were 0 false positives, where a false positive is defined as a non-fall that is misidentified as a fall. There were 9 false negatives, where a false negative is a fall that was not detected. Of the 9 falls not detected, 5 of them occurred when the subject fell to the right. Table 5 shows the results of each fall for each of the 18 subjects. Table 6 shows the fall directions attempted by each subject and the number of falls detected and not detected.
45


Table 5. Experimental results from protocol: Fall Detected(x), Fall Detected on Second Fall X(2), Fall Not Detected(Blank
Box), Fall Not Completed (-), Correct Fall Indications in Black, Incorrect Fall Indications in Red
Fall Direction Subject 1 Subject 2 Subject B Subject 4 Subject 5 Subject 6
Indicate Fall Indication Indicate Fall Indication Indicate Fall Indication Indicate Fall Indication Indicate Fall Indication Indicate Fall Indication
Fall Backwards End on Left/Right Shoulder X Right X Right X Left - - X Left X Right
Fall Backwards X Backwards X Backwards X Backwards X Backwards X Backwards X Backwards
Fall Forward End on Left/RiRht Shoulder X Left X Right X Left X - X Right X Left
Fall Forward X Right X Forward X X Forwards X Forward X Right
Lateral Fall Left End on Stomach X Right - - X Right X Right X - X Right
Lateral Left Fall X Left X Right X X - X Left X Left
Lateral Right Fall End on Stomach X Forward - - X Right X Right X - X Right
Lateral Fall Right - - X Right X Right - - x(2) Right X Right
Fall Backwards, Land in Sitting Position, End on Back X Backwards X Backwards X Backwards X Backwards X Backwards X Backwards
4*
O'


Table 5 cont. Experimental results from protocol: Fall Detected(x), Fall Detected on Second Fall X(2), Fall Not
Detected(Blank Box), Fall Not Completed (-), Correct Fall Indications in Black, Incorrect Fall Indications in Red
Su oject 7 Subject 8 Subject 9 Subject 10 Subject 11 Subject 12
Indicate Fall Indication Indicate Fall Indication Indicate Fall Indication Indicate Fall Indication Indicate Fall Indication Indicate Fall Indication
X Left X - X Left X Right X Left X Right
X - X Backwards X Backwards X Backwards X Backwards x(2) Backwards
X Right X Left X Right X Right X Forward X Right
X Forward X Forward X Right X Right X Forward X Forward
X Forward X - X Right X Right X Right X Right
X Left X Left X Right X Left X Left X Right
X Forward X Right X Right X Right X Right X Right
x(2) Right X Right x(2) Right X Right x(2) Right X -
X Backwards X Right X Backwards X Backwards X Backwards x(2) Backwards
4^
"J


Table 5 cont. Experimental results from protocol: Fall Detected(x), Fall Detected on Second Fall X(2), Fall Not
Detected(Blank Box), Fall Not Completed (-), Correct Fall Indications in Black, Incorrect Fall Indications in Red
Subject 13 Sub ect 14 Subject 15 Subject 16 Subj ect 17 Subject 18
Indicate Fall Indication Indicate Fall Indication Indicate Fall Indication Indicate Fall Indication Indicate Fall Indication Indicate Fall Indication
x(2) Left X Right X Left X Right X Left X
X Backwards X Backwards X X Backwards X Backwards X Backwards
X Left X Left X X Right X Right X Right
X Right X Forward X Right X Right X Right X Right
X Right X Right X Right X Forward X Right X
X Left X Left X Right X Right X Left
X Right X Right X Right X Right X Right X Right
X Right X Right X Left X Right X Backwards
X Backwards X Right X Backwards X Backwards X Right X Backwards
4^
CO


Table 6. Fall directions attempted by each subject and the number of falls detected and not detected.
Fall Direction Total # Falls Attempted # Falls -Detected # Falls Not Detected
Fall backwards end on left/right shoulder 18 17 1
Fall backwards 19 18 1
Fall forward end on left/right shoulder 18 18 0
Fall forward 18 18 0
Lateral fall left end on stomach 17 17 0
Lateral fall left 18 17 1
Lateral fall right end forward 17 17 0
Lateral fall right 20 15 5
Fall backwards, land on butt, end on back 19 18 1
The results from the subjects had a sensitivity, specificity, and accuracy of 95%, 100%, and 96% respectively, seen in Table 7.
Table 7. Number of falls correctly identified (True Positives), Activities of Daily Living not detected as fall (True Negative), Number of Activities of Daily Living detected as falls (False Positives), Number of falls not detected as falls (False Negatives), and the sensitivity, specificity, and accuracy.
True Positives 155
True Negatives 72
False Positives 0
False Negatives 9
Sensitivity 0.95
Specificity 1
Accuracy 0.96
49


The reason the fall detection device did not detect 8 of the 9 falls was due to
either the pitch or roll value being outside of the threshold values seen in the table below.
The lateral fall right completed by Subject 3 was not detected despite all values being
above the threshold values. The reason for this may have been because only one set of
pitch and roll values were above the threshold and the algorithm may have missed this.
Table 8. The nine falls that were not detected and a description of why it was not detected.
Subject and Fall Direction Reason Fall Was Not Triggered
Subject 3 Lateral Fall Right All Values Above Their Respective Thresholds
Subject 5 Lateral Fall Right Pitch Subject 7 Lateral Fall Right Roll< Roll Threshold, Pitch> Pitch Threshold
Subject 10 Fall Backwards, Land on Butt, End on Back RolKRoll Threshold, Pitch> Pitch Threshold
Subject 9 Fall Right Pitch>Pitch Threshold
Subject 10 Fall Backwards RolKRoll Threshold, Pitch>Pitch Threshold
Subject 11 Fall Backwards End Left Pitch Subject 14 Lateral Fall Right Pitch>Pitch Threshold
Subject 15 Lateral Fall Left Pitch The fall detection device had 93 correctly identified fall directions and 43 incorrect identifications, as seen in Table 9. Overall, it was 66% accurate in determining which direction the individual fell. Lateral Fall Right End on Stomach and Lateral Fall Left End on Stomach had the highest number of misidentified fall directions. The fall detection device indicated a lateral fall to the right for 45 of the 49 total misidentified fall directions as seen in Table 10.
50


Table 9. The fall direction and the number of directions correctly identified for each fall type.
Fall Direction # Falls -Direction Correctly Identified
Fall Backwards End Left/Right 14
Fall Backwards 16
Fall Forward End Left/Right 14
Fall Forward 8
Lateral Fall Left End Forward 2
Lateral Fall Left 10
Lateral Fall Right End Forward 2
Lateral Fall Right 12
Fall Backwards, Land on Butt, End on Back 15
Table 10. The number of fall directions that were misidentified for each fall type and the direction that was indicated for each misidentification.
Fall Direction the Sensor was Supposed to Indicate # Of Fall Directions Misidentified Fall Direction the Sensor Indicated for Each Misidentification
Fall Backwards End Left 2 2 Right
Fall Backwards End Right 0 NA
Fall Backwards 0 NA
Fall Forward End Left 2 1 Left and 1 Forward
Fall Forward End Right 0 NA
Fall Forward 9 9 Right
Lateral Fall Left End Forward 12 12 Right
Lateral Fall Left 5 5 Right
Lateral Fall Right End Forward 14 14 Right
Lateral Fall Right 2 1 Backward and 1 Left
Fall Backwards, Land on Butt, End on Back 3 3 Right
he algorithm was written so that it could detect whether the person had fallen and whether they ended laterally to the right, laterally to the left, on their back, or on their
51


stomach. It was split in these four directions because each direction had a unique range of RMS, Altitude, Pitch, and Roll values. Creating different threshold ranges for RMS, Altitude, Pitch, and Roll for each direction leveraged this. This can be seen in the distinctive graph patterns for each fall direction (Figures 14-17). The visual representation from a subject ending on a left shoulder is seen below in Figure 14. The four graphs show the output of the Roll (Figure 14a), Pitch( Figure 14b), Altitude (Figure 14c), and RMS (Figure 14d) over the period before the fall, the impact, aftermath and posture phases. These are pointed out in the figure. When a subject ends on the left shoulder the RMS spikes, the altitude decreases, the roll value increases between -32 deg and 15 deg, and the pitch value decreases between -149 deg and 71 deg. This pattern is reflected throughout the rest of the study sample.
Figure 14. Example roll, pitch, altitude, and RMS data from a fall onto the left shoulder.
52


The visual representation of the data from a subject ending on a right shoulder is shown in Figure 15. Figure 15 shows the output of the Roll (Figure 15a), Pitch (Figure 15b), Altitude (Figure 15c), and RMS (Figure 15d) over the period before the fall, the impact, aftermath, and posture phases. These are pointed out in the figure. When an individual ends on the right shoulder the RMS value spikes, the altitude decreases, the roll value increases between -25 deg and 17 deg, and the pitch value decreases -27 deg and 120 deg. This pattern is reflected throughout the entire sample.
(b)
Impact
Aftermath
Time (ms)
Time (ms)
Figure 15. Example roll, pitch, altitude, and RMS data from a fall onto the right
shoulder.
The visual representation from a subject ending on the back is seen below in Figure 16. Figure 16 shows the output of the Roll (Figure 16a), Pitch (Figure 16b), Altitude (Figure 16c), and RMS (Figure 16d) over the period before the fall, the impact,
53


aftermath, and posture phases. These are pointed out in the figure. When a subject ends on their back the RMS spikes, altitude decreases, the roll value increases between -10 deg and 7 deg, and the pitch value decreases between -230 deg and -18 deg. This pattern is reflected throughout the rest of the study sample.
Figure 16. Example roll, pitch, altitude, and RMS data from a fall onto the back.
The visual representation from a subject ending on the stomach is seen below in Figure 17. Figure 17 shows the output of the Roll (Figure 17a), Pitch (Figure 17b), Altitude (Figure 17c), and RMS (Figure 17d) over the period before the fall, the impact, aftermath, and posture phases. These are pointed out in the figure. When a subject ends on their forward the RMS spikes, altitude decreases, the roll value increases between -26 deg and 3 deg, and the pitch value decreases between -10 deg and 39 deg. This pattern is reflected throughout the rest of the study sample.
54


(a)
0
-20
cn
Q
-60
-80
70
75 80 85
Time (ms)
90
70
(d)
75 80 85
Time (ms)
90
75 80 85
Time (ms)
70
75 80 85 90
Time (ms)
Figure 17. Example roll, pitch, altitude, and RMS data from a fall onto the stomach.
The pitch and roll values that were recorded after the RMS threshold was triggered for detected falls and falls that went undetected, were collected. They were then sorted by fall direction in order to determine the range of pitch and roll values that occurred during testing. The averaged pitch and roll values for falls where the subject ended on their left shoulder are shown in Figure 18. Each fall that ended with the subject on their left shoulder had an average roll value that was extremely close to one another. The average pitch value was close for two of the falls and within one standard deviation of the third.
55


Table 11. Resulting average, standard deviation, and range of pitch and roll values for each type of fall
Fall Bac End o Shoi kwards n Left Ider Fall Forward End on Left Shoulder Lateral Left Fall Fall Fc End o Shot rward Right jlder Fall Bac End o Shot kwards Right jlder Lateral Fall Right Lateral Fall Left End on Stomach Lateral Right Fall End on Stomach Fall Forward Fall Bac Land ir Positior kwards. Sitting End on Fall Backwards
Roll Pitch Roll Pitch Roll Pitch Roll Pitch Roll Pitch Roll Pitch Roll Pitch Roll Pitch Roll Pitch Roll Pitch Roll Pitch
Average -17.07 -95.20 -7.44 -54.89 -8.31 -38.76 -1.64 74.07 -11.38 62.38 -3.96 46.79 -5.67 3.30 -10.09 12.27 -11.57 14.57 -6.09 -148.36 -1.66 -125.07
STDV 19.02 45.02 20.90 74.48 24.18 110.39 25.69 33.43 30.37 47.88 21.55 73.97 29.78 17.55 22.74 29.15 15.35 25.16 18.11 61.48 8.76 106.20
Smallest Value in Range -36.00 -140.22 -28.34 -129.37 -32.49 -149.15 -27.34 40.64 -41.76 14.51 -25.50 -27.17 -35.45 -14.25 -32.83 -16.87 -26.92 -10.59 -24.20 -209.84 -10.41 -231.27
Largest Value in Range 2.00 -50.18 13.45 19.59 15.87 71.63 24.05 107.50 18.99 110.26 17.59 120.76 24.11 20.84 12.65 41.42 3.78 39.72 12.02 -86.89 7.10 -18.87
cn
On


100.00
50.00
0.00
u -50.00 kb
0J
Q
-100.00
-150.00
-200.00
I
O
3 fa
t-H 1=
& w
3P 3 V £-
g* sT
a s
Q.
o
^ w
3
S3.
1
r rj r
CD P p
a> c^-
V n p
3 o
O E p r CD a>
CD |-1 a*
ffl 3
3
S3.
O
3
A Roll
X Pitch
Figure 18. Average pitch and roll values for falls that ended with the subject on their left
shoulder with standard deviation error bars.
The average pitch and roll values for falls that ended with the subject on their
right shoulder were all within 20 deg of each other (Figure 19). The average roll values
were once again closer together than the average pitch values.
57


140.00
A Roll
X Pitch
Figure 19. Average pitch and roll values for falls that ended with the subject on their right shoulder with standard deviation error bars.
The average pitch and roll values for falls that ended with the subject on their
stomach were both within 20 deg for each fall (Figure 20). The average roll value was
within 6 deg of each other whereas the average pitch values were within 11 deg.
58


Degrees ()
50.00
m r w r n
& 3 3
^ a S
§ 3 § 3 o 3 *K
05 n 05 n O
S e* S £ 05
3 r g a tr St> 3 a £ a. S* OQ ^ tr *K O 3 3 o
tr
A Roll
X Pitch
Figure 20. Average pitch and roll values for falls that ended with the subject on their stomach with standard deviation error bars.
The average pitch and roll values for falls where the subject ended on their
back had a close cluster for both values (Figure 21).
59


50.00
0.00
-50.00
(/)
$ -100.00 k*
on
V
Q
-150.00
-200.00
-250.00
T
JL 1 i L

N ) ' <


td ti r
fa O O E2 3-
fa fa 3 !=3 S3.
td
3 fa ^ O
cn sv
CT. fa
a 3 5
g0" S*
o
3
ffl
3
3j
P
td
3
O
sf
3
Roll
X Pitch
Figure 21. Average pitch and roll values for falls that ended with the subject on their
back with standard deviation error bars.
Mock Device Study
Of the 9 subjects that participated in the study, all 8 were able to wear the mock device for the full 3 days. One of the subjects did refuse to have the mock device placed on them the first morning due to multiple irritations and frustrations that they were experiencing. When the subject was re-approached later in the day after they had calmed down, they agreed to wear it. During the three days the subjects wore the mock device it stayed on all 3 nights for 1 subject, and fell off 1 out of the 3 nights for 3 subjects. It fell off 2 out of the 3 nights for 3 subjects, and 3 out of 3 nights for 0 subjects. Only 1
60


subject took the mock device off while wearing it. Table 12 shows the days that the mock device stayed on, the times the residents took it off, and the times it fell off at night for each subject. The NA entries in the table were because the carestaff did not have time to place the device on the subject that day.
61


Table 12. Total number of times each subject had the mock device fall off at night, took the mock device off during the day, refused to wear it, and wore it throughout the night.
Subject # Current Ambulatory Status Notes about mock device Total # of Nights it Fell Off Subject Total # of Nights it Stayed On Subject Total # of Times Subject Took Device Off Total # of Times Subject Refused to Wear It
Day 1 Day 2 Day 3
1 Walks with walker if he remembers Stayed on Fell off at night Stayed on 1 2 0 0
2 Walks on his own Stayed on Fell off at night Fell off at night 2 1 0 0
3 Walks on her own Fell off at night Fell off at night Fell off at night 3 0 0 0
4 Cane/walker/walks on her own Refused to wear it on first attempt, stayed on after second attempt Fell off at night Fell off at night 2 1 0 1
5 Walker Fell off at night Stayed on NA 1 1 0 0
6 Cane Stayed on Stayed on Stayed on 0 3 0 0
7 Walks on her own Fell off at night Stayed on Stayed on 1 2 0 0
8 Walks on her own Stayed on Took it off NA 0 1 1 0
9 Walks on her own Fell off at night Fell off at night NA 2 0 0 0
Overall Total 12 11 1 1
CT\


Final Device Testing
Of the 9 subjects that participated, 7 were able to wear the FDS for the full 3 days. During the three days the 7 subjects wore the FDS no one fell while wearing it. Additionally, none of the activities of daily living were detected as falls while the subjects wore the FDS. Subject 5 was not approached on the first day to wear the device due to the irritations and frustrations they were having with the day, and then on the second and third day they said they did not want to wear it. Subject 8 took the device off twice during the first day due to irritation. The subjected stated that it was itchy and bothering her. Due to this reaction they were not asked to wear the device again during day two or three. While wearing the real device, it did not fall off at all for 2 subjects, and it fell off 1 out of the 3 nights for 1 subject. It fell off 2 out of the 3 nights for 3 subjects, and 3 out of 3 nights for 0 subjects. It fell off during the day for only one subject. Table 13, shown below, shows the days that the mock device stayed, the times the residents took it off, and the times it fell off at night for each subject. Table 14, shown below, shows the results from the system usability survey. Overall, the average score from the Chelsea Place Memory Care staff was a 57/100. The NA entries in the table were because the carestaff did not have time to place the device on the subject that day
63


Table 13. Total number of times each subject had the real fall device fall off at night, took the real fall device off during the
Subject# Current Ambulatory Status Notes about real device Total # of Nights it Fell Off Subject Total # of Nights it Stayed On Subject Total # of Times Subject Took Device Off Total # of Times Subject Refused to Wear It Total # of Times It Fell Off Subject During the Day
Day 1 Day 2 Day 3
1 Walks with walker if he remembers Stayed on Fell off at night Fell off at night 2 1 0 0 0
2 Walks on his own Fell off at night Stayed on Stayed on 1 2 0 0 0
3 Walks on her own Fell off at night Fell off at night NA 2 0 0 0 0
4 Cane/walker/walks on her own Fell off at night Fell off at night Took it off 2 0 1 0 0
5 Walker Refused on first attempt Refused on first attempt Refused on first attempt 0 0 0 3 0
6 Cane Stayed on Stayed on NA 0 2 0 0 0
7 Walks on her own Stayed on Stayed on Stayed on 0 3 0 0 0
8 Walks on her own Irritation, replaced and then took it off NA NA 0 0 2 0 0
9 Walks on her own Fell off during the day Stayed on 0 1 0 0 1
Overall Total 7 r 9 3 3 1
Os
4^


Table 14. Chelsea Place Memory Care staff system usability survey results.
Subject # System Usability Score
1 67.5
2 42.5
3 55
4 50
5 72.6
Average 57.52
65


CHAPTER IV
DISCUSSION AND FUTURE WORK Fall Detection System Development
The fall detection algorithm was split into three thresholds based on the success of this method in Pierloni et al., 2015s research. This method was successful because each parameter changes when a person falls, and using more than one of those parameters improves the likelihood of accurately detecting a fall. The final threshold, pitch and roll, was broken down into four different directions to help improve the identification of how a person landed and where an injury may be. The range of values for each direction were based on the range of values the sensor outputted when it was resting in each of the four directions.
The battery life of the system was also an important consideration when designing the device. Multiple steps to improve the battery life, while keeping the design lightweight and compact, were taken. For example, the data sampling rate was decreased to 0.25 s, and the FDS only connected to the WiFi and MQTT after all four thresholds were met. These changes nearly doubled the overall battery life of the FDS. Two different lithium polymer batteries that could power the device for more than a day were identified. One was a smaller battery (23 g, 34mm x 62mm x 5mm) that would last for a little over a day, and the other was a larger battery (52g, 51mm x 65mm x 8mm) that would last for 3 days. The caretakers at Chelsea Place Memory Care were asked which one they would prefer. They stated that they wanted to check on the device every morning during testing, thus they preferred the smaller battery that lasted for a day.
66


The smartphone application IFTT was the last step in the fall detection system. The fall detection system takes 2 to 5 minutes to send the fall notification. The reason for this is because there is a delay in how often it checks the fall3 feed on the Adafruit MQTT. In the future, a smartphone application needs to be developed specifically for this system so that the notifications can be sent quickly and efficiently
Young Healthy Volunteer Study
This study helped validate the algorithm used for the Fall Detection System. The results of 95% sensitivity, 100% specificity, and 96% accuracy, while not as high of sensitivity and accuracy as Pierloni et al., 2015, validated that the fall detection system is able to detect a large number of falls by people of varying height and age. The 100% specificity is critical because it proves that the device would not send a false alarm to an already very busy caretaker. Of the nine falls that were not detected, 5 of them were lateral falls to the right, two were falls that ended on the back, and one was a fall backwards ending on the left shoulder. Each of the falls that were not detected met the RMS, aftermath, and altitude threshold. The reason 8 of the 9 falls were not detected was because the pitch and roll values were outside of the threshold. The last fall that was not triggered met all of the thresholds, but did not send a final notification. This may of happened because all four thresholds were only met for one data point, which could have caused the algorithm to miss the fall indication. The algorithm can be improved to detect an even higher percentage of falls by altering the threshold ranges to reflect the range of values that indicated a fall in the test data.
The directional indication of the fall detection system was 66% accurate. The low percentage value can be attributed to the large number of falls that were classified as falls
67


ending on the right shoulder. Of the 43 falls that were misidentified, 39 of them were misidentified as lateral right fall. One reason for this, is that the pitch and roll threshold for lateral right fall had the largest range. This large range resulted in significant overlap with the pitch and roll threshold ranges for the other three fall directions. Another reason for the low accuracy percentage is that when the subjects fell onto the matt they would land in the correct orientation, bounce into a different orientation, and then stand up quickly. The multiple movements immediately after the fall may have affected the algorithms ability to determine the correct orientation. Further examination of the values from the recorded falls could provide more accurate values for the pitch and roll thresholds for future testing.
The unique pitch and roll values for each fall direction resulted in distinctive graph patterns. The graphs also show the three phases of a fall: impact, aftermath and posture, and altitude change. The altitude change is consistently between 0.5 and 2 m.
The impact value, while easily visible by the sharp RMS increase, changes based on how fast the person fell. The aftermath phase that follows is characterized by the flat acceleration line around lg. The values of pitch and roll values are the only values that change based on the direction of the fall.
The pitch and roll values that were recorded after the RMS threshold was triggered for detected falls and falls that went undetected, were collected. They were then sorted by fall direction in order to determine the range of pitch and roll values that occurred during testing. These values were averaged, and a standard deviation was found. The standard deviation was found in order to determine more accurate fall direction threshold ranges. The resulting average values for the falls that ended on the left shoulder
68


were all within a standard deviation of each other. The same can be said for falls that ended on the right shoulder, stomach, and back. This shows that falls ending in the same position resulted in similar pitch and roll values. These averaged values and standard deviations provide new, more accurate, pitch and roll ranges for each of the four fall directions.
Mock Device and Final Device Study
The mock device stayed on for approximately half of the nights it was worn, and fell off for the other half. The real device stayed on for a little under half of the nights it was worn. One subject wore the mock device for two days. On the third day they took off the mock device, and then when they transitioned to wearing the real device they took that one off as well. The subject stated that it was itching her. Due to this irritation she was not asked to wear it any longer. A second subject also took the real device off on the third day they were wearing it. These results suggest that this placement is a good location for most subjects, and that it will not bother them. It also shows that for some individuals an addition of a Band-Aid will cause too much irritation. Creating a thinner device with a smaller battery that can fit within the Band-Aid could solve the irritation that some subjects felt. It could also solve the issue of the device falling off every other night.
The device only fell off one subject during the day because the adhesive would no longer stick to the person. This could have been because the subject only bathes once every three to four days and their skin was too oily. Further research into better adhesive quality should be done. Additionally, one of the sensors was lost after a resident took off the device when a caretaker was not around. This device was found a day later after the
69


subjects room was cleaned. This is a common occurrence with many objects in a memory care facility, and needs to be taken into consideration when designing a device for this population in the future. One subject either refused to wear the device or were not approached to wear the device due to the mood that they were in three days in a row. These results highlight the heightened daily irritation and frustration this population experiences during daily life due to the difficulties dementia presents. It also plays a huge factor in whether the individual is willing to wear a new device or not. This suggests that a device that needs to be placed on the individual every few days may be too obtrusive for certain people with dementia.
Once the testing was complete, staff at Chelsea Place Memory Care that interacted with the fall detection system were asked to complete the system usability survey. The average score was 57/100. The survey results show that the carestaff found the system easy to use and learn, not complex or cumbersome, and were somewhat confident while using it. Despite this, most would not like to use this system. The reason for this was because they had to continually check to make sure it was there, and replace it in the morning if it fell off. It added additional steps to their already very busy morning schedule. These are important results because if the carestaff does not find the device useful it will not be used properly. If they system was able to stay on for multiple days, did not require the battery to be changed so often, and was smaller it may be more usable for the carestaff.
Future Fall Detection System Studies
In the future this system needs to continually refined, and the accuracy of the four threshold values needs to be improved. This can happen by examining the current
70


threshold test results, and by using a machine learning technique. Once it is able to accurately detect which way the person fell, it will help caretakers know where to look for injuries and help the users receive faster care. The algorithm also needs to be improved so that it can detect syncope falls. A syncope fall is defined as a fall where the user collapses against a wall, but ends up sitting. The altitude data from these tests needs to be looked at further as well. This needs to be completed in order to look at the variability of altitude changes during activities of daily living, and to determine what would be a more acceptable threshold value. The data between the four activities of daily living versus a fall need to be compared further as well in order to better determine the differences between the two and improve the threshold values. This device also needs to be more compact so that the device and the battery can all fit easily within a Band-Aid. One way this can be achieved is by combining the two boards onto one surface mount printed circuit board and finding a smaller battery. This, in addition to a slightly larger Band-Aid, may also help the device stay on the residents throughout the night. Furthermore, improvements need to be made to the battery life. Ideally, the device should last for up to a week so that caretakers are not constantly changing the battery. This can be accomplished by finding a smaller battery with higher amperage, and by finding ways to conserve power within the algorithm. This will also allow longer studies to be completed with the residents. The longer testing period will increase the likelihood of a resident falling while wearing the FDS and provide data to compare to the young adult study. In future iterations, a smartphone application needs to be specifically designed for this system. This will allow the system to be independent of a third party application, send notifications faster, and work more efficiently within Anthem memory cares current
71


system. A possible GPS locater or RFID tag would also be advantageous for this device so that it would not be lost. The device should also be waterproofed in the future to make it much more user friendly. This would also be beneficial so that the device could detect any falls that would occur in the shower or bathroom.
72


CHAPTER V
CONCLUSION
A real-time, accurate, wireless fall detection device that detects falls that occur from a standing position by a resident with dementia in a memory care facility was developed. It combined a microcontroller, 10-Degree of Freedom IMU sensor, battery, WiFi, and a smartphone application. Once this was created, healthy individuals between the ages of 20-56 complete 9 different falls and 4 activities of daily living. The calculated sensitivity, specificity, and accuracy for the device were 95%, 100%, and 96% respectively. These results validated the algorithm that was developed for the device. Once this was complete, residents at Chelsea Place Memory Care were recruited to wear the mock device for 3 days, and then the real device for an additional 3 days. This was to determine whether this device was properly designed for this population. It was found that this device is currently too large and falls off at night as a result. It was also found that the placement of the device on the lower shoulder caused little irritation or frustration to the residents throughout the day.
73


REFERENCES
1) Ortman, J., Velkoff, V., & Hogan, H. (2014). An Aging Nation: The Older Population in the United States. Retrieved from https://www.census.gov/prod/2014pubs/p25-1140.pdf.
2) Wiener, Joshua, and Jane Tilly. "Population Ageing in the United States of America: Implications for Public Programmes." Int. J. Epidemiology2002. 776-81. Vol. 31
3) 2016 ALZHEIMER'S DISEASE FACTS AND FIGURES. (2016). Retrieved from http://www.alz.org/facts/overview.asp
4) Tabet, N., & Nelson, L. (2015). Slowing the progression of Alzheimers disease; what works? (Vol. 23, pp. 193-209). Ageing Research Reviews.
5) Waite, L. (2015). Treatment for Alzheimer's disease: has anything changed? (Vol. 38, pp. 60-63). Australian Prescriber.
6) Barton, C., Ketelle, R., Merrilees, J., & Miller, B. (2016). Non-pharmacological Management of Behavioral Symptoms in Frontotemporal and Other Dementias (Vol. 16): Current Neurology and Neuroscience Reports.
7) Campdell, J., Robertson, C., Gardner, M., Norton, R., Tilyard, M., & Buchner, D. (1997). Randomised controlled trial of a general practice programme of home based exercise to prevent falls in elderly women
8) Rubenstein, L. (2006). Falls in older people: epidemiology, risk factors and strategies for prevention (Vol. 35, pp. 37-41). age and aging.
9) Eriksson, S., Gustafson, Y., & Lundin-Olsson, L. (2008). Risk factors for falls in people with and without a diagnose of dementia living in residential care facilities: A prospective study (Vol. 46, pp. 293-306). Archives of Gerontology and Geriatrics.
10) Doom, C. v., Gruber-Baldini, A. L., Zimmerman, S., Hebei, J. R., Port, C., Baumgarten, M., Magaziner, J. (2003). Dementia as a Risk Factor for Falls and Fall Injuries Among Nursing Home Residents. Retrieved from http://www.medscape.com/viewarticle/460913
11) Tinetti, M. E., Liu, W. L., & Claus, E. B. (1993). Predictors and prognosis of inability to get up after falls among elderly persons (Vol. 269, pp. 65-70). J. Am. Med. Assoc.
12) Maki, E. B. (1997). Gait Changes in Older Adults: Predictors of Falls or Indicators of Fear? (Vol. 45, pp. 313-320): The American Geriatrics Society.
13) Gagnon, N., & Flint, A. J. (2003). Fear of Falling in the Elderly (Vol. 6, pp. 15-17). Geriatrics and Aging.
74


14) Auvinet, E., Multon, F., Saint-Amaud, A., Rousseau, J., & Meunier, J. (2011). Fall Detection With Multiple Cameras: An Occlusion-Resistant Method Based on 3-D
15) Belshaw, M., Taati, B., Snoek, J., & Mihailidis, A. (2011). Towards a single sensor passive solution for automated fall detection (pp. 1773 1776). 2011 Annual International Conference of the IEEE Engineering in Medicine and Biology Society.
16) Sherwin, S., & Winsby, M. (2010). A relational perspective on autonomy for older adults residing in nursing homes (Vol. 14, pp. 182-190): Health Expectations.
17) Gasparrini, S., Cippitelli, E., Spinsante, S., & Gambi, E. (2014). A Depth-Based Fall Detection System Using a Kinect Sensor (Vol. 14, pp. 2756-2775). Sensors.
18) Madansing, S., Thrasher, T., Layne, C., & Lee, B.-C. (2015). Smartphone based fall detection system (pp. 370 374). Control, Automation and Systems (ICCAS), 2015 15th International Conference on.
19) He, Y., Li, Y., & Bao, S.-D. (2012). Fall detection by built-in tri-accelerometer of
smartphone (pp. 184 187). Proceedings of 2012 IEEE-EMBS International Conference on Biomedical and Health Informatics.
20) Fang, S.-H., Liang, Y.-C., & Chiu, K.-M. (2012). Developing a mobile phone-based fall detection system on Android platform (pp. 143 146). Computing, Communications and Applications Conference (ComComAp), 2012.
21) Li, Y., Ho, K. C., Popescu, M., & Skubic, M. (2014). A Theoretical Study on The Placement of Microphone Arrays for Improving The Localization Accuracy of A Fall (pp. 4523-4526).
22) Li, Y., Popescu, M., Ho, K. C., & Nabelek, D. (2011). Improving Acoustic Fall Recognition by Adaptive Signal Windowing. 33rd Annual International Conference of the IEEE EMBS.
23) Li, Y., Zeng, Z., Popescu, M., & Ho, K. C. (2010). Acoustic Fall Detection Using a Circular Microphone Array (pp. 2242-2245). 32nd Annual International Conference of the IEEE EMBS.
24) Werner, F., Diermaier, J., Schmid, S., & Panek, P. (2011). Fall Detection with Distributed Floor-mounted Accelerometers
25) Chaccour, K., Darazi, R., Hajjam el Hassani, A., & Andres, E. (2015). Smart Carpet using differential piezoresistive pressure sensors for elderly fall detection (pp. 225-229).
26) Aud, M., Abbott, C., Tyrer, H. W., Neelgund, R. V., Shriniwar, U. A. B., &
Devarakonda, K. K. (2010). Smart Carpet: Developing a Sensor System to Detect Falls and Summon Assistance (pp. 8-12).
75


27) Danes, S. (2012). Design for Dementia Care: A Retrospective Look at The Woodside Place Mode (Vol. 26, pp. 221-250): Journal of Housing for the Elderly.
28) Uegami, M., Iwamoto, T., & Matsumoto, M. (2012). A Study of Detection of Trip and Fall Using Doppler Sensor on Embedded Computer (pp. 3263-3268). IEEE International Conference on Systems, Man, and Cybernetics.
29) Liu, L., Popescu, M., Skubic, M., Rantz, M., Yardibi, T., & Cuddihy, P. (2011). Automatic Fall Detection Based on Doppler Radar Motion Signature (pp. 222-225). 5th International Conference on Pervasive Computing Technologies for Healthcare (PervasiveHealth) and Workshops.
30) Liu, L., Popescu, M., Skubic, M., & Rantz, M. (2014). An Automatic Fall Detection Framework Using Data Fusion of Doppler Radar and Motion Sensor Network (pp. 5940-5943).
31) Garripoli, C., Mercuri, M., Karsmakers, P., Soh, P. J., Crupi, G., Vandenbosch, G. A. E.,
. Schreurs, D. (2015). Embedded DSP-Based Telehealth Radar System for Remote In-Door Fall Detection (Vol. 19, pp. 92-101). IEEE JOURNAL OF BIOMEDICAL AND HEALTH INFORMATICS.
32) Igual, R., Medrano, C., & Plaza, I. (2013). Challenges, issues and trends in fall detection systems (Vol. 12, pp. 1-24): BioMedical Engineering OnLine.
33) Bourke, A. K., Ven, P. v. d., Gamble, M., OConnor, R., Murphy, K., Bogan, E.,... Nelson, J. (2010). Applications of Waist Segment Kinematic Measurement using Accelerometry for an Autonomous Fall-detection System during Continuous Activities (pp. 198-203). Signals and Systems Conference (ISSC 2010).
34) Chen, D., Feng, W., Zhang, Y., Li, X., & Wang, T. (2011). A Wearable Wireless Fall Detection System with Accelerators (pp. 2259-2263). International Conference on Robotics and Biomimetics.
35) Barton, C., Ketelle, R., Merrilees, J., & Miller, B. (2016). Non-pharmacological Management of Behavioral Symptoms in Frontotemporal and Other Dementias (Vol. 16): Current Neurology and Neuroscience Reports.
36) Bharucha, J. S., Anand, V., Forlizzi, J., Dew, M. A., Reynolds III, F. C., Stevens, S., et al. (2009). Intelligent Assistive Technology Applications to Dementia Care: Current Capabilities, Limitations, and Future Challenges (Vol. 17, pp. 88-104): The American Journal of Geriatric Psychiatry.
37) Igual, R., Medrano, C., & Plaza, I. (2013). Challenges, issues and trends in fall detection systems (Vol. 12, pp. 1-24): BioMedical Engineering OnLine.
76


38) Dumitrache, M., & Pasca, S. (2013). Fall Detection Algorithm Based on Triaxial Accelerometer Data (pp. 1-4). E-Health and Bioengineering Conference (EHB).
39) Lan, C.-C., Hsueh, Y.-H., & Hu, R.-Y. (2012). Real-Time Fall Detecting System Using a Tri-axial Accelerometer for Home Care (pp. 1077-1080). International Conference on Biomedical Engineering and Biotechnology.
40) Tolkiehn, M., Atallah, L., Lo, B., & Yang, G.-Z. (2011). Direction Sensitive Fall Detection Using a Triaxial Accelerometer and a Barometric Pressure Sensor (pp. 369-372). 33rd Annual International Conference of the IEEE EMBS.
41) Bianchi, F., Redmond, S. J., Narayanan,M R., Cerutti, S., & Lovell, N. H. (2010a). Barometric Pressure and Triaxial Accelerometry-Based Falls Event Detection (Vol. 18, pp. 619-627). IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING.
42) Wang, C., Narayanan, M. R., Lord, S. R., Redmond, S. J., & Lovell, N. H. (2014). A Low-power Fall Detection Algorithm Based on Triaxial Acceleration and Barometric Pressure (pp. 570-573). Conference Proceeding IEEE Engineering Medical Biology Society.
43) Pierleoni, P., Belli, A., Palma, L., Pellegrini, M., Pemini, L., & Valenti, S. (2015). A High Reliability Wearable Device for Elderly Fall Detection (Vol. 15, pp. 4544-4553). SENSORS
44) Onishi, J., Suzuki, Y., Umegaki, H., Endo, H., Kawamura, T., Imaizumi, M., et al.
(2006). Behavioral, psychological and physical symptoms in group homes for older adults with dementia (pp. 75-86): International Psychogeriatric Association.
45) Rubenstein, L. (2006). Falls in older people: epidemiology, risk factors and strategies for prevention (Vol. 35, pp. 37-41). age and aging.
46) Karantonis, D. M., Narayanan, M. R., Mathie, M., Lovell, N. H., & Celler, B. G. (2006). Implementation of a real-time human movement classifier using a triaxial accelerometer for ambulatory monitoring (Vol. 10, pp. 156-167). Trans. Inf. Technol. Biomed.
77


APPENDIX
A. Arduino Fall Detection Code
//include "Adafmit_MQTT.h"
//include "Adafmit_MQTT_Client.h" include
//include
//include
//include
//include
//include
//include
//include
//include
//include
//define AIOSERVER "io.adafruit.com"
//define AIO SERVERPORT 8883 // use 1883 8883 for SSL
//define AIOUSERNAME "chrimack"
//define AIO KEY Md9f866da3bd74bf39ec7d83155e0d342"
/* Assign a unique ID to the sensors */
AdafruitlODOF dof = Adafruit_10DOF();
Adafruit_LSM303_Accel_Unified accel = Adafruit_LSM303_Accel_Unified(30301); Adafruit_LSM303_Mag_Unified mag = Adafruit_LSM303_Mag_Unified(30302); Adafruit_BMP085_Unified bmp = Adafruit_BMP085_Unified(18001);
/* Update this with the correct SLP for accurate altitude measurements */ float seaLevelPressure = SENSORS PRESSURE SEALEVELHPA;
//float insTemp, desTemp, outTemp;
//charfall[100] = "\{\"Roll\":orientation.roll, \"Pitch\":orientation.pitch,
\"Altitude\":altitude 1 \, \"RMS\":rmsl\, \"Fall\":A, \"Time\":timel\}";
char fall2[1000] = "\{\"Roll2\":R, \"Pitch2\":P, \"Altitude2\":A\, \"RMS2\":M\,
\"Fall\":F\, \"Time\":G\}";
// Update these with values suitable for your network.
//const char* ssid = "HoosierHouse"; const char* ssid = "Guest";
//const char* ssid = "Mackenzie Christensen";
//const char* password = "gopackgo"; const char* password = "willowBrookl";
//const char* ssid = "More Beer, More Fun";
//const char* password = "DG860A336DC2";
78


const char* mqtt server = "ml 1.cloudmqtt.com";
const char* id="abcd";
const char* usr="wfkyabjj";
const char* pass="6PlMAjsE01m8";
WiFiClient client;
Adafruit MQTT Client mqtt(&client, AIOSERVER, AIOSERVERPORT, AIOU SERNAME, AIO KEY);
AdafruitMQTTPublish fall3 = Adafruit_MQTT_Publish(&mqtt,
AIOU SERNAME 7f/fall3");
WiFiClient wifiClient;
Pub SubClient mqttclient(wifiClient);
*******/ void initSensors()
{
if(!accel.begin())
{
/* There was a problem detecting the LSM303 ... check your connections */ Serial.println(F("Ooops, noLSM303 detected ... Check your wiring!")); fall3.publish("Accelorometer not connected"); while(l);
}
if(!mag.begin())
{
/* There was a problem detecting the LSM303 ... check your connections */ Serial.println("Ooops, noLSM303 detected ... Check your wiring!"); fall3.publish("Magnetometer not connected"); while(l);
}
if(!bmp.begin())
{
/* There was a problem detecting the BMP 180 ... check your connections */ Serial.println("Ooops, no BMP180 detected ... Check your wiring!"); fall3.publish("BMP180 not connected"); while(l);
}
}
79


long lastMsg = 0; char msg[50]; int value = 0; float altitude 1; float altitude2; float altitude3; float altitude4; float rms; float rmsl;
int myrms[1000]={ };
float myalt[1000]={ };
int mypitch[1000]={ };
int myroll[ 1000]={ };
char* myfall[1000]={ };
int mytime[1000]={ };
int sample = 0;
int J=0;
int R=0;
int M=0;
int A=0;
char* F=0;
int P=0;
int G=0;
int roll=0;
int pitch=0;
void MQTT_connect(); void setup() {
int Atl=2.5; //g int TtA=l; //s int AltA=.72; //g int AutA=1.28; //g int Otp=50; //degrees int Ttp=l; //s int rms2=0; int rms3=0; int rms4=0; int sample=0;
pinMode(BUILTIN_LED, OUTPUT); // Initialize the BUILTIN LED pin as an output
Seri al.begin(l 15200); setup_wifi();
mqttclient. setServer(mqtt_server, 26954);
MQTT_connect();
}
80



void setup_wifi() {
delay(lO);
Serial.println();
Serial.println(ssid);
WiFi.begin(ssid, password);
while (WiFi.statusO != WLCONNECTED) { delay(50);
Serial.print(".");
}
Serial.println("");
Serial.println(WiFi.localIP());
/* Initialise the sensors */ initSensors();
MQTT_connect();
}
jjsf.************************** MQTT
void MQTT_connect()
{
int8_t ret;
if (mqtt. connected!)) { return;
}
uint8_t retries = 3;
while ((ret = mqtt.connect()) != 0) {
Serial.println(mqtt.connectErrorString(ret));
mqtt.disconnect();
delay(lO); // wait 5 seconds
retries;
if (retries == 0) { while (1);
}
}
}
81


II'Initial Loop
void loop() { WiFi.forceSleepBegin(); delay(lOO);
sensors_event_t accel_event; sensors_event_t mag_event; sensors_event_t bmp_event; sensors vec t orientation;
/* Calculate pitch and roll from the raw accelerometer data */ accel ,getEvent(&accel_event); float
rms=abs((sqrt((sq(accel_event.acceleration.x))+(sq(accel_event.acceleration.y))+(sq( accel_event.acceleration.z))))-10); accel. getEvent(& accel_event);
if (dof. accel GetOri entati on(&accel_event, & orientation))
{
/* 'orientation1 should have valid .roll and .pitch fields */ ori entati on. rol 1=ori entati on. r ol 1; ori entati on. pitch=ori entati on .pitch; rms=rms;
}
//* Calculate the altitude using the barometric pressure sensor */ bmp. getEvent(&bmp_event); if (bmpevent. pressure)
{
/* Get ambient temperature in C */ float temperature;
bmp.getTemperature(&temperature);
/* Convert atmospheric pressure, SLP and temp to altitude */ float seaLevelPressure=SENSORS_PRESSURE SEALEVELHPA; altitudel=(bmp.pressureToAltitude(seaLevelPressure, bmp_event.pressure, temperature));
altitude 1 =altitude 1; unsigned long timel=millis(); int my2darray[100][4]={ };
82


char* f-'Normal";
my rm s [ sampl e]=rm s;
myalt[sample]=altitudel;
myroll[sample]=orientation.roll;
my pitch [ sampl e]=ori entati on. pitch;
myfall[sample]=f;
mytime[sample]=timel;
}
sample++; if (sample>=400){ sampl e=0;
}
delay(250);
p^|J clctCCtlOn loOp
if (rms > 4)
{
for (int i=0; i <= 9; i++){
sensors_event_t accel_event; sensors_event_t mag_event; sensors_event_t bmp_event; sensors vec t orientation;
/* Calculate pitch and roll from the raw accelerometer data */ accel ,getEvent(&accel_event); float
rmsl=abs((sqrt((sq(accel_event.acceleration.x))+(sq(accel_event.acceleration.y))+(sq (accelevent. accel erati on. z))))-10);
accel. getEvent(& accel_event);
if (dof. accel GetOri entati on(&accel_event, & orientation))
{
/* 'orientation' should have valid .roll and .pitch fields */
83


on entati on. rol 1=ori entati on. r ol 1; int roll=orientation.roll;
ori entati on. pitch=ori entati on .pitch; int pitch=ori entati on. pitch;
rmsl=rmsl;
}
/* Calculate the heading using the magnetometer */
/* Calculate the altitude using the barometric pressure sensor */ bmp. getEvent(&bmp_event); if (bmpevent. pressure)
{
/* Get ambient temperature in C */ float temperature;
bmp.getTemperature(&temperature);
/* Convert atmospheric pressure, SLP and temp to altitude */
altitude2=(bmp.pressureToAltitude(seaLevelPressure,
bmp_event.pressure,
temperature));
altitude2=altitude2;
unsigned long timel=millis(); int my2darray[100][4]={ }; char* q="RMS Triggered";
my rm s [ sampl e]=rm s 1;
myalt[sample]=altitude2;
myroll[sample]=orientation.roll;
my pitch [ sampl e]=ori entati on. pitch;
myfall[sample]=q;
mytime[sample]=timel;
sample++;
}
84


if (sample>=400){ sample=0;
}
delay (10);
if (l<=rmsl<=2) //needs to check next rms event
{
float J = altitude2 altitude 1; if (abs(J)>0.4)
{
delay(lO);
if (orientation.roll>=-25 && orientation.roll<=15 && orientation.pitch>=-25 && orientation.pitch<= 15)
{
for (int o=0; o <= 100; o++){ sensors_event_t accel_event; sensors_event_t mag_event; sensors_event_t bmp_event; sensors vec t orientation;
/* Calculate pitch and roll from the raw accelerometer data */ accel ,getEvent(&accel_event); float
rmsl=abs((sqrt((sq(accel_event.acceleration.x))+(sq(accel_event.acceleration.y))+(sq (accelevent. accel erati on. z))))-10);
accel. getEvent(& accel_event);
if (dof. accel GetOri entati on(&accel_event, & orientation))
{
/* 'orientation' should have valid .roll and .pitch fields */
ori entati on. rol 1=ori entati on. r ol 1; roll=orientation.roll;
ori entati on. pitch=ori entati on .pitch; pitch=ori entati on. pitch;
rmsl=rmsl;
// Serial.print(F("; "));
}
/* Calculate the heading using the magnetometer */
85


*/ //* Calculate the altitude using the barometric pressure sensor bmp. getEvent(&bmp_event); if (bmp event, pressure) { /* Get ambient temperature in C */ float temperature; bmp.getTemperature(&temperature);
*/ /* Convert atmospheric pressure, SLP and temp to altitude
timel=millis(); altitude3=(bmp.pressureToAltitude(seaLevelPressure, bmp_event.pressure, temperature)); altitude3=altitude3; unsigned long int my2darray[100][4]={ }; char* n="Forward"; my rm s [ sampl e]=rm s 1; myalt[sample]=altitude3; myroll[sample]=orientation.roll; my pitch [ sampl e]=ori entati on. pitch; myfall[sample]=n; mytime[sample]=timel; sample++; }
//Serial. print(sample++); if (sample>=400){ sampl e=0; } } WiFi.forceSleepWake(); delay(lO);
86


Serial. println();
Serial.println(ssid);
WiFi.begin(ssid, password);
while (WiFi.statusO != WLCONNECTED) { delay(50);
}
Serial.println("");
Serial.println(WiFi.localIP());
MQTTconnectQ;
while (Imqttclient. connected!)) {
// Attempt to connect if (mqttclient.connected,usr,pass)) {
// Once connected, publish an announcement... mqttclient.publish("topic", "hello world");
II... and resubscribe mqttclient.subscribe("inTopic");
} else {
Serial .print(mqttclient. state());
// Wait 5 seconds before retrying delay(5000);
}}
initSensors(); fall3.publish("l"); fall3.publish("5"); for (int 1=0; 1 < 400; l++){
StaticJsonBuffer<200> j sonBuffer; JsonObject& data =j sonBuffer. createObject(); float M=(myrms[l]);
data["RMS"]=double_with_n_digits(M,4); float A=(myalt[l]);
data[" Altitude"]=doubl e_with_n_digits( A, 6); //Serial.print(A);
R=(myroll[l]);
data["Roll"]=R;
P=(mypitch[l]);
87


data["Pitch"]=P;
F=(myfall[l]);
data["Fall"]=F;
G=(mytime[l]);
data["Time"]=G;
data.printTo(fall2, sizeof(fall2));
mqttclient.publish("fall2",fall2,2);
//Serial.print(O);
}
fall3.publish("0"); }
else if (orientation.roll>=-70 && orientation.roll<=50 && orientation.pitch>=-25 && orientation.pitch<=l 10)
{
//mqttclient.publish("fall3",''Fall to the Right Detected",2);
//fall3.publish("2");
for (int p=0; p <= 100; p++){
sensors_event_t accel_event;
sensors_event_t mag_event;
sensors_event_t bmp_event;
sensors vec t orientation;
/* Calculate pitch and roll from the raw accelerometer data */ accel ,getEvent(&accel_event); float
rmsl=abs((sqrt((sq(accel_event.acceleration.x))+(sq(accel_event.acceleration.y))+(sq (accelevent. accel erati on. z))))-10);
accel. getEvent(& accel_event);
if (dof. accel GetOri entati on(&accel_event, & orientation))
{
/* 'orientation' should have valid .roll and .pitch fields */
ori entati on. rol 1=ori entati on. r ol 1; roll=orientation.roll;
ori entati on. pitch=ori entati on .pitch; pitch=ori entati on. pitch;
rmsl=rmsl;
// Serial.print(F("; "));
}
88


*/ /* Calculate the heading using the magnetometer */ //* Calculate the altitude using the barometric pressure sensor bmp. getEvent(&bmp_event); if (bmp event, pressure) { /* Get ambient temperature in C */ float temperature; bmp.getTemperature(&temperature);
*/ /* Convert atmospheric pressure, SLP and temp to altitude
timel=millis(); altitude4=(bmp.pressureToAltitude(seaLevelPressure, unsigned long int my2darray[100][4]={ }; char* w="Right"; my rm s [ sampl e]=rm s 1; myalt[sample]=altitude4; myroll[sample]=orientation.roll; my pitch [ sampl e]=ori entati on. pitch; myfall[sample]=w; mytime[sample]=timel; sample++; } if (sample>=400){ sampl e=0; } }
WiFi.forceSleepWake(); delay(lO); Serial.println(); //Serial.print("Connecting to "); Serial.println(ssid);
89


WiFi.begin(ssid, password);
while (WiFi.statusO != WLCONNECTED) { delay(50);
//Serial.print(".");
}
Serial.println("");
//Serial.println("WiFi connected 2");
//Serial.println("IP address: "); Serial.println(WiFi.localIP()); MQTTconnectQ;
while (!mqttclient.connected()) {
// Attempt to connect if (mqttclient.connected,usr,pass)) {
// Once connected, publish an announcement... mqttclient.publish("topic", "hello world");
II... and resubscribe mqttclient.subscribe("inTopic");
} else {
//Serial.print("failed, rc=");
Serial .print(mqttclient. state());
//Serial.println(" try again in 5 seconds");
// Wait 5 seconds before retrying delay(5000);
}}
initSensors();
fall3.publish("2"); fall3.publish("5"); for (int y=0; y < 400; y++){
StaticJsonBuffer<200> j sonBuffer; JsonObject& data =j sonBuffer. createObject(); float M=(myrms[y]);
data["RMS"]=double_with_n_digits(M,4); float A=(myalt[y]);
data[" Altitude"]=doubl e_with_n_digits( A, 6); Serial.print(A);
R=(myroll[y]);
90


data["Roll"]=R;
P=(mypitch[y]);
data["Pitch"]=P;
F=(myfall[y]);
data["Fall"]=F;
G=(mytime[y]);
data["Time"]=G;
data.printTo(fall2, sizeof(fall2)); mqttclient.publish("fall2",fall2,2);
}
fall3.publish("0"); }
else if (orientation.roll>=-20 && orientation.roll<=20 && orientation.pitch>=-190 && orientation.pitch<=-150)
{
for (int v=0; v <= 100; v++){ sensors_event_t accel_event; sensors_event_t mag_event; sensors_event_t bmp_event; sensors vec t orientation;
/* Calculate pitch and roll from the raw accelerometer data */ accel ,getEvent(&accel_event); float
rmsl=abs((sqrt((sq(accel_event.acceleration.x))+(sq(accel_event.acceleration.y))+(sq (accelevent. accel erati on. z))))-10);
accel. getEvent(& accel_event);
if (dof. accel GetOri entati on(&accel_event, & orientation))
{
/* 'orientation' should have valid .roll and .pitch fields */
ori entati on. rol 1=ori entati on. r ol 1; roll=orientation.roll;
ori entati on. pitch=ori entati on .pitch; pitch=ori entati on. pitch;
rmsl=rmsl;
// Serial.print(F("; "));
}
91


/* Calculate the heading using the magnetometer */
*/ //* Calculate the altitude using the barometric pressure sensor bmp. getEvent(&bmp_event); if (bmp event, pressure) { /* Get ambient temperature in C */ float temperature; bmp.getTemperature(&temperature);
*/ /* Convert atmospheric pressure, SLP and temp to altitude
timel=millis(); float altitude4=(bmp.pressureToAltitude(seaLevelPressure, bmp_event.pressure, temperature)); altitude4=altitude4; unsigned long int my2darray[100][4]={ }; char* a="Backwards"; my rm s [ sampl e]=rm s 1; myalt[sample]=altitude4; myroll[sample]=orientation.roll; my pitch [ sampl e]=ori entati on. pitch; myfall[sample]=a; mytime[sample]=timel; sample++; } if (sample>=400){ sampl e=0; } } WiFi.forceSleepWakeQ;
92


delay(lO);
Serial. println();
//Serial.print("Connecting to "); Serial.println(ssid);
WiFi.begin(ssid, password);
while (WiFi.statusO != WLCONNECTED) { delay(50);
//Serial.print(".");
}
Serial.println("");
Serial.println(WiFi.localIP()); MQTTconnectQ;
while (!mqttclient.connected()) {
// Attempt to connect if (mqttclient.connected,usr,pass)) {
// Once connected, publish an announcement... mqttclient.publish("topic", "hello world");
II... and resubscribe mqttclient.subscribe("inTopic");
} else {
//Serial.print("failed, rc=");
Serial .print(mqttclient. state());
//Serial.println(" try again in 5 seconds");
// Wait 5 seconds before retrying delay(5000);
}}
initSensors();
fall3.publish("3");
fall3.publish("5");
for (int h=0; h < 400; h++){
StaticJsonBuffer<200> j sonBuffer; JsonObject& data =j sonBuffer. createObject(); float M=(myrms[h]);
data["RMS"]=double_with_n_digits(M,4); float A=(myalt[h]);
data[" Altitude"]=doubl e_with_n_digits( A, 6);
93


Full Text
xml version 1.0 encoding ISO-8859-1
DISS_submission publishing_option 0 embargo_code third_party_search Y
DISS_authorship
DISS_author type primary
DISS_name
DISS_surname Christensen
DISS_fname Mackenzie
DISS_middle
DISS_suffix
DISS_affiliation University of Colorado at Denver
DISS_contact current
DISS_contact_effdt 04/23/2017
DISS_phone_fax P
DISS_cntry_cd 1
DISS_area_code 608
DISS_phone_num 235-6694
DISS_phone_ext
DISS_address
DISS_addrline N66 W38533 N Woodlake Circle
DISS_city Oconomowoc
DISS_st WI
DISS_pcode 53066
DISS_country US
DISS_email christensen.mackenzie@gmail.com
future
04/23/2017
1
608
235-6694
N66 W38533 N Woodlake Circle
Oconomowoc
WI
53066
US
christensen.mackenzie@gmail.com
DISS_citizenship US
DISS_description page_count 167 masters external_id http:dissertations.umi.comucdenver:10850 apply_for_copyright no
DISS_title DETECTING FALLS IN RESIDENTS WITH DEMENTIA IN A MEMORY CARE FACILITY USING A REAL-TIME WIRELESS FALL DETECTION DEVICE: A PILOT STUDY
DISS_dates
DISS_comp_date 2017
DISS_accept_date 01/01/2017
DISS_degree M.S.
DISS_institution
DISS_inst_code 0765
DISS_inst_name University of Colorado Denver
DISS_inst_contact Bioengineering
DISS_processing_code N
DISS_advisor
Bodine
Cathy
DISS_cmte_member
Sliker
Levin
Lammers
Steven
DISS_categorization
DISS_category
DISS_cat_code 0541
DISS_cat_desc Biomedical engineering
DISS_keyword Dementia Care, Elderly Care, Fall Detection System, Fall Sensor
DISS_language en
DISS_content
DISS_abstract
DISS_para The collaboration between the Neurocognitive Technology for Aging Lab and Anthem Memory Care identified a significant need for a real-time, comfortable, non-obtrusive fall detection device that can be worn by individuals with dementia. There is currently no fall detection technology built specifically for this population despite the fact that they are nearly twice as likely to fall compared to someone without dementia. A fall detection device was designed to fill this need. It consisted of a small microcontroller and a 10-Degree of Freedom IMU sensor. The device detects the direction of a fall, and sends a notification to a smartphone application. Eighteen young healthy individuals tested the fall detection device. Each subject completed 9 different falls into a matt and 4 activities of daily living (ADL). It resulted in a specificity, sensitivity, and accuracy of 100%, 95%, and 96% respectively and was 66% accurate when identifying the direction of the fall. Once this was complete, 9 residents of Anthem’s Chelsea Place Memory Care wore a mock fall detection device for 3 days, and then wore the real fall detection device for another 3 days. This was completed to determine that the device was comfortable and unobtrusive to the daily lives of the residents.
DISS_supp_abstract
DISS_binary PDF Christensen_ucdenver_0765N_10850.pdf
DISS_restriction
DISS_repository
DISS_version 2011-11-08 15:37:33
DISS_agreement_decision_date 2017-04-23 22:23:19
DISS_acceptance 1
DISS_delayed_release
DISS_access_option



PAGE 1

DETECTING FALLS IN RESIDENTS WITH DEMENTIA IN A MEMORY CARE FACILITY USING A REAL TIME WIRELESS FALL DETECTION DEVICE : A PILOT STUDY b y MACKENZIE L. CHRISTENSEN Bachelor of Science, Rose Hulman Institute of Technology, 2015 A thesis submitted to the Fa culty of the Graduate School of the University of Colorado in partial fulfillment of the requirements for the degree of Master of Science Bioengineering Program 201 7

PAGE 2

! "" This thesis for the Master of Science degree by Mackenzie L. Christensen has be en approved for the Bioengineering Program b y Cathy Bodine, Chair Levin Sliker Steven Lammers Date: May 13, 2017

PAGE 3

! """ Christensen, Mackenzie L. (M.S. Bioengineering Program ) Detecting Falls in Residents with Dementia in a Memory Care Facility U sing a Real Time Wireless Fall Detection Device: A Pilot Study Thesis directed by Associate Professor Cathy Bodine ABSTRACT The collaboration between the Neurocognitive Technology for Aging Lab and Anthem Memory Care identified a significant need for a real time, comfortable, non obtrusive fall detection device that can be worn by individuals with dementia. There is currently no fall detection technology built specifically for this population despite the fact that they are nearly twice as likely to fal l compared to someone without dementia. A fall detection device was designed to fill this need. It consisted of a small microcontroller and a 10 Degree of Freedom IMU sensor The device detects the direction of a fall and sends a notification to a smartph one app lication Eighteen young healthy individuals tested the fall detection device. Each subject completed 9 different falls into a matt and 4 activities of daily living (ADL). It resulted in a specificity, sensitivity, and accuracy of 100%, 95%, and 96% respectively and was 66% accurate when identifying the direction of the fall. Once this was complete 9 residents of Anthem's Chelsea Place Memory Care wore a mock fall detection device for 3 days and then wore the real fall detection device for another 3 days. This was completed to determine that the device was comfortable and unobtrusive to the daily lives of the residents The form and content of this abstract are approved. I recommend its publication. Approved: Cathy Bodine

PAGE 4

! "# ACKNOWLEDGEMENTS Thank y ou so much to the many inspiring professors I have had, my committee members, friends, and my amazing parents for helping, supporting, and encouraging me throughout my educational endeavors. I could not have made it here without you!

PAGE 5

! # TABLE OF CONTENTS CHA PTER I. INTRODUCTION ................................ ................................ ................................ ........... 1 Increasing Population of People over 65 with Dementia ................................ .... 1 Current Fall Detection Technology Used ................................ ........................... 3 Video Cameras ................................ ................................ ................................ ............ 3 Smartphones ................................ ................................ ................................ ................ 5 Microphones ................................ ................................ ................................ ............... 7 Floor Sensors ................................ ................................ ................................ .............. 8 Doppler Radar ................................ ................................ ................................ ........... 10 Impact of Technology to Detect or Predict Falls ................................ .............. 18 Partnership with Anthem Memory Care ................................ ........................... 19 II. SPECIFIC AIMS ................................ ................................ ................................ .......... 20 III. MATERIALS AND METHODS ................................ ................................ ................ 21 Fall Detection System Development ................................ ................................ 21 Young Healthy Volunteer Study ................................ ................................ ....... 30 Mock Device Study ................................ ................................ ........................... 34 Final Device Testing ................................ ................................ ......................... 38 IV. RESULTS ................................ ................................ ................................ ................... 44 Fall Detection System Developm ent ................................ ................................ 44 Young Healthy Volunteer Study ................................ ................................ ....... 45 Mock Device Study ................................ ................................ ........................... 60 Final Device Testing ................................ ................................ ......................... 63

PAGE 6

! #" V. DISCUSSION AND FUTURE WORK ................................ ................................ ....... 66 Fall Detection System Development ................................ ................................ 66 Young Healthy Volunteer Study ................................ ................................ ....... 67 Mock Device and Final Device Study ................................ .............................. 69 Future Fall Detection System Studies ................................ ............................... 70 VI. CONCLUSION ................................ ................................ ................................ ........... 73 REFERENCES ................................ ................................ ................................ .................. 74 APPENDIX A. Arduino Fall Detection Code ................................ ................................ ....... 78 B. Python Fall Data Collection Code ................................ ................................ 98 C. Arduino Code to Test Battery Life Constantly Connected to Wifi ............ 100 D. Arduino Code to Test Battery Life Only Connecting to WiFi When Fall Triggered ................................ ................................ ................................ ........ 106 E. Python Code Collecting Sensor Data ................................ ......................... 112 F. Python Code Collecting Battery Data ................................ ......................... 114 G. Fall Backward End on Left Shoulder Graphs for Young ........................... 116 H. Fall Forward End on Left Shoulder Graphs ................................ ............... 118 I. Lateral Left Fall Graphs ................................ ................................ .............. 120 J. Fall Forward End on Right Shoulder Graphs ................................ .............. 124 K. Fall Backwards End on Right Shoulder Graphs ................................ ........ 127 L. Lateral Fall R ight Graphs ................................ ................................ ........... 129 M. Lateral Fall Left End on Stomach Graphs ................................ ................. 134 N. Lateral Right Fall End on Stomach Graphs ................................ ............... 139

PAGE 7

! #"" O. Fall Forward Graphs ................................ ................................ .................. 144 P. Fall Backwards, Land in Sitting Position, End on Back Graphs ................ 149 Q. Fall Backwards Graphs ................................ ................................ .............. 154 R. System Usability Survey ................................ ................................ ............ 159

PAGE 8

! $ CHAPTER I INTRODUCTION Increasi ng Population of People over 65 with Dementia According to the United States Census Bureau, "In 2050, the population aged 65 and over is projected to be 83.7 million, almost double its estimated population of 43.1 million in 2012," ( 1 ). This projection wil l be one of the biggest challenges facing the nation, and the world, in coming years. One of the most concerning issues for this aging population is that each individual will have their own medical concerns that will have to be addressed and paid for. One of the most prevalent diseases in the aging population is dementia ( 2 ). According to the World Health Organization The number of people living with dementia worldwide is currently estimated at 47.5 million and is projected to increase to 75.6 million by 2030" ( 3 ). Dementia is the decline of memory or other thinking skills such as memory loss, judgment, language, complex motor skills, and other intellectual function due to permanent damage or death of the brain's nerve cells or neurons ( 3 ). These symptoms are often severe enough to reduce a person's ability to perform everyday activities ( 3 ) There are many different types of dementia such as lewy body, vascular, frontotemporal, and Alzheimer's, which is the most common. Alzheimer's is a debilitating disea se and is the fifth leading cause of death in the United States for those age 65 and older ( 3 ). It is the only top 10 cause of death in America today that cannot be prevented, cured, or slowed ( 3 ). The only treatment options available are monitoring of blo od pressure, healthy diet, regular exercise, cognitive stimulation therapy, and social networks ( 4 ). Another treatment is prescription drugs. Of the approved drugs, such as cholinesterase and

PAGE 9

! % memantine, it has been shown that the overall risks do not warra nt using them and that they do not slow the progression of dementia ( 5 ). Recent research has also shown that the best way to treat individuals with dementia is with customized care using creative management of physical therapies, environmental factors, an d behavioral therapy (6). Increased Risk of Falling Due to Dementia Most treatments for persons with dementia are aimed at delaying the onset of common symptoms. However, increased probability of falling is one symptom that has few preventative treatment options. Physical therapy and environmental preventative measures can reduce the risk of falling ( 7 ). This can include something as simple as short daily exercises of walking ( 7 ). Other common measures are de cluttering rooms, increasing lighting, and addi ng assistive devices, such as canes, walkers, wheelchairs, and fall pads ( 8 ). Despite these efforts, the unadjusted fall rate was 4.05 per year for residents in a nursing home diagnosed with dementia, compared with 2.33 falls per year for residents without dementia ( 9 ). This is mostly due to symptoms such as impaired judgment, gait instability, visual spatial perception, and a decreased ability to recognize and avoid hazards ( 10 ). For individuals with dementia who experience frequent falls there is a concer n whether anyone will be present when a fall happens. The results of a fall can be as minor as a bruise or as serious as a traumatic brain injury, or even death ( 11 ). Every elderly individual who falls develops a fear of falling again and this fear leads t o the individual falling more often ( 12 ). Researchers call this phenomenon "space phobia" or a person's loss of confidence in balance and walking. Community based epidemiologic studies have found that 21 61% of elderly people experience some degree of fear of falling even if

PAGE 10

! & they have never previously fallen. Furthermore, of the elderly individuals who are afraid of falling 70% say that they avoid certain activities due to this fear. This can often lead to depression and anxiety ( 13 ). Current Fall Detectio n Technology Used Given the increased fall rate in people with dementia researchers are working to develop reliable technologies to detect falls. Currently, they can be classified into a few different groupings: Video Cameras, Smartphones, Microphones, Fl oor Sensors, Doppler, and accelerometers/gyroscope/barometer sensor combinations. In the following paragraphs each of these will be discussed. Video Cameras Significant research has been completed on the use of video camera's to detect falls due to the ir ability to monitor an individual without being obtrusive to their daily life. They are also very accurate, as can be seen in Auvinet, Multon, Saint Arnaud, Rousseau, & Meunier, 2011 ( 1 4) and Belshaw, Taati, Snoek & Mihailidis, 2011 ( 15 ). Belshaw, et a l., 2011 utilize d an artificially intelligent camera based system that automatically detects if a person within the field of view has fallen. The system uses a consumer grade camera with a wide angle lens. A machine learning technique allow es the system to classify whether a person has fallen or not at a high accuracy rate. It takes into account lighting, environment and the presence of moving objects. The testing resulted in a true positive rate of 92% and a false positive rate of 5%. The large advancement in this study was that the system was able to handle multiple moving objects in a room and determine when one of those objects fell. It also accounted for large lighting changes throughout the day ( 15 ).

PAGE 11

! Auvinet et al's., 2011 study provided another advan cement by utilizing a new method of reconstructing the 3 D shape of people. Falls were detected by analyzing the volume distribution along the vertical axis, and an alarm triggered when the major part of this distribution wa s abnormally near the floor duri ng a predefined period of time. This method was tested with videos of healthy subjects who performed 24 realistic scenarios showing 22 fall events and 24 activities of daily living. A 99.7% sensitivity and specificity was achieved with four cameras or more The sensitivity decreased to 80.6% when they decreased the number of cameras to three. The researchers conclude d that this study compares well with other literature, but acknowledge that falls that end up on a couch or chair could be missed due to the bo dy not being close to the floor. They also acknowledge that the only way to use this system in a multi room setting would require setting up 4 to 6 cameras in each room ( 1 4). Despite the advantages of using camera systems, there were quite a few problems with the technology. The most important problem was that it required the recording of people, which means their privacy was being violated. This is not acceptable in a residential memory care community that values their resident s anonymity. Sherwin & Win sby (2010) discuss how one aspect of autonomy for residents is their privacy and this is an important concern for many residents and families ( 16 ). To solve this problem, research groups have begun using depth based camera's in order to maintain people's privacy, such as Gasparrini, Cippitelli, Spinsante, & Gambi, 2014 ( 17 ). Depth based cameras allow a person to show up on the camera without any identifiable features It also identifies the different objects in the room, and can tell how far away they are from the camera. All of these features allow the camera to create an accurate picture of the

PAGE 12

! ( room and identify where the person is while maintaining their privacy. Gasparrini et al., ( 2014 ) used a Microsoft Kinect to use a frame by frame analysis of the r oom. The system extracts elements and classifies them as objects or as a person. Once the person was identified, a tracking algorithm was put into place in order to monitor their movements. The tracking algorithm allowed the system to handle the individual 's interactions with stationary objects. When the person was detected near the floor, the system predict ed they ha d fallen ( 17 ). Despite advances in privacy with this system the researchers stated there are still two major problems that have not been solv ed using depth cameras These include the number of cameras required to cover an entire room ; the computational power required to sort through all of the data ; and the inability of camera systems to handle more than one person in the room. Each of these challenges present a large hurdle for camera systems to overcome in order to accurately detect falls and make this technology useful Smartphones Other technologies under exploration are accelerometers, gyroscopes, and magnotometers built into today's sm art phones. These sensors detect movements of the user and are implemented in many health apps that are found on phones today. Quite a few research groups have begun investigating this technology to determine its accuracy and usability and will be discusse d in the following paragraphs ( 18 19 20 ). In a study completed by He, Li, & Bao, ( 2012 ), the group utilized the built in tri accelerometer to collect information about body movement ( 19 ). They determined that body motion could be classified into five di fferent patterns: vertical activity, lying, sitting or static standing, horizontal activity and fall. When this system detects a fall it sends a

PAGE 13

! ) text message to a pre selected group of people. The text message includes time, GPS coordinate s and a Google map of the suspected fall location. During testing for this system the phone was worn at the waist of the user T he researchers mentioned that it presented a problem because most people do not keep their phones mounted at the waist. Instead, many people ke ep their phones in their pants or shirt pocket. They also mentioned that in future studies they will need to solve the problem of adapting the technology to daily wear and that it needs to work when the user places the i r phone somewhere besides their wais t. In a study completed by Madansing, Thrasher, Layne, & Lee, ( 2015 ) the smart phone system was divided into three phases: basic architecture, analysis, and communication. The basic architecture is the flow diagram of how the smartphone system works. The accelerometer, gyroscope, and barometer in the phone output data is collected, analyzed, and communicated via an email, text, or a phone call. The analysis portion describe d how the collected data from the phone is then analyzed and interpreted in order to determine if a person has fallen or not. The final phase is communication In this step the system notifies the corresponding individual of the fall over email, text, or a phone call. This study went on to discuss the different locations that the smartpho ne could be placed for fall detection, such as the waist or chest. They also stated that the components found in a smartphone that could be used for fall detection are temperature g a uges touch sensors, cameras, and ambient light sensors. They then analyze d 25 different studies that used smartphones for fall detection. What they found was that each of these studies used only simulated falls, an android was the most accurate fall detecting system, and the accelerometer was the most used sensor followed by th e gyroscope.

PAGE 14

! This study also found quite a few challenges with the smartphone technology itself. For example, the quality of sensors inside phones were not high enough to produce consistently accurate data. The energy consumption and battery life of thi s technology was very high and required a lot of time to recharge. The final challenge was placement and usability, because carrying a smartphone in a single pocket or position is difficult to do. The researchers mentioned that while this technology is inc redibly convenient and easily accessed on a phone, it is not feasible for the elderly population nor is it practical for individuals with dementia. They discussed how the current aging population did not grow up with cell phones and most of them do not own a cell phone or know how to use one Furthermore, they are not used to carrying it around every where and for elderly individuals who have memory loss there is a very good chance they will not remember where their phone is or to keep it in one position al l day ( 18 ). This is the main reason why using the accelerometers, gyroscopes, and magnotometers inside phones is not practical for this population. Microphones Another type of technology that is an option for fall detection is sound sensors and microphon es. These have become more and more popular because unlike cameras, they are able to maintain people's privacy, and can be discretely installed in any room of a house or building. In a study completed by Li, Ho, Popescu, & Skubic, ( 2014 ) they installed 8 m icrophones on a circular wooden board and hung it on a wall ( 21 ). The system was comprised of three different processing steps: sound source localization, increase the signal to noise ratio, and fall recognition. This study consisted of 12 falls and 12 non falls and found that analyzing sound at the beginning of the signal resulted in a

PAGE 15

! + more accurate identification of a fall or non fall. But a study by Li, Popescu, Ho, & Nabelek, 2011 ( 22 ) found that one limiting factor was that it operated under the assum ption that there wa s only one person in the room. Additionally, it require d a lot of microphones set up around the room in order to filter out the extra sounds, such as phone calls, typing, or a TV. Studies such as Li, Zeng, Popescu, & Ho, ( 2010 ) ( 23 ) foun d that it took an 8 microphone linear array to most accurately detect all of the extra reverberating sound. The same group also found that it takes at least a 2 microphone array in order to somewhat accurately locate sounds in a room ( 21 ). Based on the stu dies described above it would be difficult to use this technology at memory care facilities with many rooms, increased noise levels, and multiple people moving in a room at once. Floor Sensors ,-.!/0110-!2.-203!4564!"2!.627!40!82.!6-9!562!:..-!/0-2"9.3.9!" -!.;9.3;7! /0118-"4".2!"2!6!<3.2283.!2.-203=!>5.3.!56#.!:..-!6!?.@!3.2.63/5!A308<2B!@5"/5! @";;!:.!9.2/3":.9!"-!45.!?0;;0@"-A!<636A36<52B!4564!56#.!82.9!45.2.!2.-2032!:7! .1:.99"-A!45.1!"-! /63<.42!C %' B! %( B! %) D=!>5.7!"-4.A364.!18;4"<;.!<3.2283.!2.-2032! @"45"-! 6!2E863.!/63<.4!<"./.=!F!?6;;!"2!9.4./4 .9 0-!45.!/63<.4!2E863.2!@5.-!6!?03/.!"2! :."-A!6<<;".9!40!%!03!103.!<3.2283.!2.-2032!"-!6!2"-A;.! 2E863.= One study completed by Aud, Abbott, Tyrer, Neelgund, Shriniwar, & Devarakonda, 2010 developed a smart carpet th at consisted of an array of four sensors that are battery independent and placed in the carpet. To test the carpet they created a 10ft long prototype of the carpet Eleven volunteers between the ages of 20 and 60 walk ed on the carpet and answer ed a few que stions about it None of the volunteers could identify a difference between the regular carpet and the smart carpet. The carpet did have a number

PAGE 16

! G of false positives and false negatives regarding whether a person was stepping on the sensor or not during the trial. The carpet was able to determine a number of characteristics about a person's ga it such as velocity and frequency, but no actual falls were tested on th e carpet The research team concluded that the smart carpet was able to detect human gait charac teristics and that detection of a fall, which they would deem as a larger number of sensors triggered than when it was just a footstep, wa s feasible ( 26 ). Despite these conclusions, some concerns about this technology are that it has not been tasted with a ctual falls I t also has only been tested with one person in a room on one piece of carpet, and does not take into account a room with multiple wheelchair users. In another study completed by Chaccour, Darazi, Hajjam, El Hassani, & Andres, 2015, piezo res istive pressure sensors were used in the carpets ( 25 ). The system wa s divided into two parts: the first wa s the pressure sensors that detected the signal and the second was the processing of that signal. This system utilized a threshold based algorithm to detect a person's falls and then sent an SMS alarm notification in case of a fall event. In the experiment for this study they are examining falls and daily activities. For the experimental falls the volunteer was instructed to fall on his side and cover at least three different sensors. Elders were not considered in this experiment to avoid injury. The result was that the sensors triggered an alarm 8/9 times and did not signal an alarm when the volunteers ran, walked, jumped, or sat on the carpet ( 25 ). This was a very limited experimental study due to the fact that the volunteers had to position themselves in a specific way in order to ensure the alarm went off. In these studies the sensors have been somewhat accurate in determining if someone has fall en or not, but none of have been used in a real world community or

PAGE 17

! $H tested what will happen when there is more than one person walking on the carpet Another problem is that many dementia facilities prefer hard wood floors, especially in bedrooms and bathro oms where falls are most likely, because it is easier to clean up common incontinence accidents ( 27 ). So while this technology has a lot of benefits and potential, it is not entirely appropriate for the population of people with dementia. Doppler Radar A method that has become very popular in the past few years for detecting falls is using Doppler radar because it maintains an individual's privacy, is accurate, and can be discretely placed around the room so people do not know it is there. Originally th is technique did not start out very precise, Uegami, Iwamoto, & Matsumoto, 2012 found it was around 88 96% accurate ( 28 ). This study used a system that acquired and digitized the sensor output, processed the sensor data, and then transmitted the results t o their computer. The movements that were tested were tripping and falling, walking, shaking arm and hand, using a cell phone, sitting on a chair, and, standing without moving. Each movement was tested 20 different times. The system resulted in 20 activiti es that were falsely identified as falls. Liu, Popescu, Skubic, Rantz, Yardibi, & Cuddiby, 2011 study improved these results with a proposed fall detection technique that utilized two pulse Doppler range control radars ( 29 ). They estimated the velocitie s of subjects within the detection range and tried to recognize a fall based on its Doppler signature and velocity. They used a pilot dataset of 109 falls and 341 non fall human activities, and obtained an accuracy of 91% and 97% for detecting falls and no n falls respectively. But as the technology continued to develop Liu, Popescu, Skubic, & Rantz, 2014 ( 30 ) completed another study that found

PAGE 18

! $$ the best location for these sensors was on the ceiling, and that it was even more accurate when combined with a mot ion sensor. Additionally, Garripoli, Mercuri, Karsmakers, Soh, Crupi, Vandenbosch, & Schreurs, 2015 ( 31 ) was able to use this technology to detect 100% of falls with complete accuracy. In this study the system that was tested consisted of a sensor, combi ning radar, computational, wireless communication, and a base station for data processing. The experiment was performed in a 5m x 5m room with one volunteer at a time. There was furniture in the room to mimic a real world room setting. The sensor was fixed to a wall at a height of 1.25m while the base station was positioned 4m away from the sensor. The system was validated by having 16 volunteers, 14 males and 2 females, simulate 65 fall events. Each volunteer was monitored for 5 minutes total including bef ore and after the fall. Additionally, 40 random walking activities, 30 activities of sitting down or standing up, and 20 random movements such as closing windows and moving a chair were also observed. The results were that there were no false positives, bu t they did find that the farther away the subject was from the Doppler radar the more likely they were to get lost in the noise of the objects around them. While this wa s a significant advancement, the biggest remaining problem with Doppler radar is that i t cannot detect if a person has fallen when the person is obstructed by objects such as a bed or couch Accelerometer/Barometer/Gyroscope Sensors The most researched type of technology for fall detection are accelerometers, gyroscopes, barometers, and mag netometers ( 32 ). Each of these sensors have been used either separately or together in order to collect data about the user's body position, detect if they have fallen or not, and then send an alert that the user has fallen ( 33 34 35 36 ).

PAGE 19

! $% One of the bigge st challenges for using these types of sensors in fall detection is determining the best location to place them on the user. Most researchers have determined that the most accurate location is the waist, trunk or head due to less movement in these areas, c ompared to feet, legs, or wrists ( 37 ). Early research used a tri accelerometer to detect individual body motion in the X, Y, and Z axes Bourke et al., 2010 ( 33 ) performed a study using a single accelerometer at the waist in conjunction with a threshold a lgorithm. This threshold algorithm used specific thresholds set for the kinematic and angular data collected, and when a specific combination of thresholds was reached, it would signal a fall had occurred. In this study, multiple threshold algorithms were created in order to determine which one was the most accurate. The device was placed on 10 young healthy subjects who performed 240 falls from various starting points and directions and 120 activities of daily living. The device was also placed on 10 elder ly healthy subjects who performed 240 scripted and 52.4 hours of continuous unscripted normal activities. The results showed that the algorithm combining the parameters of impact, posture, and velocity achieved the lowest false positive rate of .94 false p ositive per day with a sensitivity of 94.6% and a specificity of 100%. Sensitivity was defined as the percentage of falls that were correctly identified as falls, and specificity as the number of daily activities that were accurately classified. Overall, t his study was successful and the use of elderly individuals was very valuable, but it may not be practical for a large community of patients with dementia. The false positive rate is very high for a larger population, and this device was worn at the user's waist making it obtrusive for this sensitive population.

PAGE 20

! $& Another group that used a similar threshold based fall detection algorithm was Dumitrache & Pasca, 2013 ( 38 ). This group used the algorithm to compare the thresholds of six different parameters dur ing 34 simulated falls and 200 daily activities that resulted in a sensitivity of 97.05% and a specificity of 99%. Overall, the algorithm of this study was extremely accurate. The major limitation of this study was that one person was used to complete all of the simulated falls and daily activities. Furthermore, the individual was a young man, so it was not an accurate representation of the elderly population or the population with dementia. Another problem as it relates to a population with dementia is t hat it requires the user to remember to wear the device and to place it at their waist daily. A study performed by Chen, Feng, Zhang, Li, & Wang, 2011 ( 34 ) used a single accelerometer and a new algorithm to develop a device that had more accurate sensitivi ty with more movements. In this study, the researchers performed 195 simulated falls and 75 daily activities that resulted in a sensitivity of 97% and a specificity of 100%. The accelerometer was placed in a large square box that was worn on a belt at the user's wa i st. The algorithm, which was more accurate than previous studies, measured the total sum acceleration and the tilt angle of the body during different phases of a fall event. This algorithm included three thresholds and three delay times, which he lped determine if the body motion and posture indicated a fall or not. Despite success with this algorithm, this study did not use its device on any elderly individuals, which may alter the number of falls that were detected It also still required the ind ividual to remember to wear a large obtrusive device at their waist. Lan, Hsueh, & Hu, 2012 ( 39 ) performed a study with one accelerometer and a different type of algorithm than previous studies. This study used a support vector

PAGE 21

! $' machine framework to proces s the accelerometer data. This method could only discern up to 4 different kinds of fall events and failed to recognize 5 falls out of 120, which resulted in an accuracy of 95.83%. This system was less accurate and less robust compared to the earlier studi es It also required the device to be worn in a box on the waist and was only tested on young subjects. The next evolution of this technology combined accelerometers with gyroscopes and barometers to help improve the detection of body movement and locatio n. Tolkiehn, Atallah, Lo, & Yang, 2011 (4 0 ) performed a comparative study using an accelerometer alone to using an accelerometer combined with a barometer. In this study, measurements included features of the body tilt while accelerating, the impact magnit ude, and the body tilt angle while no acceleration change occurred. When a fall was detected using these features, it was verified with the barometric pressure sensor data It determined if the pressure rose at the same time the suspected fall occurred; va lidating if the person actually fell. The study included 12 young healthy subjects, and each had to perform 13 different fall activities and 11 daily living activities. Following each simulated fall onto a mattress, the subjects were told to remain fallen for 15 25 seconds to simulate the lying down period that some elderly people experience after falling. The accelerometer alone was 81.48% accurate, 83.33% specific, and 79.08% sensitive, whereas the accelerometer and pressure sensor combined resulted in 86 .97% accurate, 85.24% specific, and 87.77% sensitive. The results from this study showed that using an accelerometer and a pressure sensor was more accurate in detecting falls than using an accelerometer alone. In a study by Bianchi, Redmonds, Narayanan, Cerutti, & Lovell, 2010 ( 41 ), a barometer was combined with an accelerometer to create a sensor that was worn at the

PAGE 22

! $( hip to detect a fall. This study performed three different tests to determine if the sensor accurately detected falls. The first test was comprised of only indoor movements and falls, the second test only outdoor simulated falls, and the third test indoor and outdoor simulated normal activities. The study used 16 different movements, 8 different fall movements, and 8 different daily activiti es. In these experiments, all simulated falls landed onto a mattress. This study included 20 young healthy volunteers. Three different types of algorithms were tested to see which one had the highest accuracy of fall detection. The first algorithm was base d on the extreme impact of a fall and movement intensity that can be obtained from calculating the signal magnitude vector (SVM). When the SVM reaches a signal higher than 1.8 G, it was considered a fall. The second algorithm was based on the first algorit hm but assume d that a fall always ends with a non standing body orientation of the wearer. Using this system, a fall is detected if the data indicates a large impact followed by a non standing body position. The third algorithm takes into consideration th e barometric pressure measurement when determining if a fall has occurred. It follows the same assumptions as the earlier algorithms but adds the change in altitude of the device at waist level when a fall occurs. This study resulted in an accuracy, sensi tivity, and specificity of 96.9%, 97.5%, and 96.5% respectively using algorithm three. This was an improvement from studies using only an accelerometer. However, fall detection devices could be improved by testing with an elderly population, continued impr ovement of accuracy, making the device smaller when worn at the hip, and performing a larger number of falls and daily activities to validate accuracy ( 41 ).

PAGE 23

! $) Another study that used both an accelerometer and a barometer was Wang, Narayanan, Lord, Redmond, & Lovell, 2014 (4 2 ), whose main goal was to create a system that not only detected falls but also used less power. This study included 20 young individuals who performed 8 simulated falls and 8 daily activities. The individuals wore the device at their wa ist, which detected falls using thresholds from both the accelerometer and the barometer. The device was low power, sampling at 5Hz and then increased to 40 Hz when the data met the threshold criteria to signal a fall. Overall, the results of the study wer e promising in that it showed 95.9% accuracy, 96.7% sensitivity, and 96.9% specificity. It also showed that the original algorithm, without changing sampling frequency, was about 1% more accurate. This study did not use elderly individuals in their sample and was still worn obtrusively at the hip. However, the sacrifice of accuracy for power was an important concept to consider since this device was worn daily. The most recent advancement in using accelerometers, gyroscopes, and barometers for fall detecti on has been the combination of the three sensors into an inertial measurement unit (IMU). These units often contain a three axis accelerometer, three axis gyroscope, and a three axis magnetometer, but some units contain additional advanced features. The IM U is an important advancement, because it makes the overall system much smaller and less obtrusive. This combination sensor has recently become more popular and frequently utilized in fall detection research. Pierleoni, Belli, Palma, Pellegrini, Pernini, & Valenti, 2015 ( 43 ), conducted a study that used this sensor and tested it against multiple algorithms. In this study, an IMU sensor that included a tri axial accelerometer, gyroscope and magnetometer with a microcontroller, Bluetooth module,

PAGE 24

! $* mass storage unit, battery, and wireless receiver in the form of a phone or computer was used. Different algorithms were used from two other previous studies that only looked at impact and posture. An additional algorithm was created that looked at the impact, aftermat h, and posture. The new algorithm extracted the acceleration and pitch and roll angles, which are rotations around the Y and X axe s respectively, and compared them to threshold values. These were selected after an accurate training process using the suppor t vector machine method was completed. This training method consisted of 50 simulated falls and 50 daily activities. The sensor was placed in a small hard box that was worn on the waist and tested on 10 volunteer subjects from the ages of 22 to 29. Each s ubject repeated 18 scenarios comprised of nine simulated falls and nine daily activities replicated three times, resulting in 540 tests. Subjects performed a simulated fall onto a mattress to ensure they were safe, and no elderly subjects were used in this test. The new algorithm performed better than all other algorithms it was compared with. It had 100% accuracy, 100% sensitivity, and 100% specificity when using one experimental protocol, and 90.37% accuracy, 80.74% sensitivity, 100% specificity when usin g a second experimental protocol. The difference in the two protocols was that the second protocol include d 2 falls where the subject end ed up sitting with their back staying vertical to the ground. The falls in the first protocol all end ed with the subjec t lying down with their back horizontal to the ground. Additionally, the algorithm took into consideration the position at the end of the fall, which indicated that this algorithm was able to more accurately detect the typology of the fall and the end posi tion of the user. Overall, this study proposed a very accurate algorithm that was able to detect most, if not all, falls and provided a good indication of the end position of the fall. The limitations to this study

PAGE 25

! $+ were that this method was not tested on a n elderly population and it ha d to be worn at the waist. The addition of a barometer to this system may further improve the location and position accuracy due to its ability to determine a change in height. Impact of Technology to Detect or Predict Falls Fall detection systems have made significant progress but few have been specifically designed for individuals with dementia or for use in a memory care facility ( 36 ). Those with dementia are very sensitive to sensory stimuli and regularly deal with memory loss, decreased eyesight, pain in joints and legs, and a decreased ability to communicate or understand others, which can result in tremendous frustration ( 44 ). A fall detection system needs to be developed that takes these health conditions into consider ation so that it can be used successfully with this population. B ecause of the complexity of caring for individuals with dementia, most families decide that the best place for their loved one is in a memory care facility with caretakers who have extensive experience ( 4 5 ). By 2050, the number of people using long term care services will likely double from 13 million in 2000 to 27 million people ( 4 5 ). This is mostly due to the increasing number aging people ( 4 5 ). One of the biggest challenges facing these fac ilities will be keeping track of the number of falls that occur ( 4 5 ). Typically, long term care facilities follow a protocol called "Standard of Care" These are a set list of guidelines that the facility must follow in order to properly care for and monit or the residents, especially if they are at an increased risk of falling. These guidelines are defined by broad requirements set by federal and state law under 42 CFR ¤483 of the Nursing Reform Act. Under this protocol, the long term care facility checks o n the residents once every hour to ensure that the resident is okay and has not fallen. Despite this procedure there are still

PAGE 26

! $G many falls that go unseen and unreported (45). Due to this there is a critical need to design a real time accurate, wireless fal l detection device to detect falls that occur from a standing position by a resident with dementia in a memory care facility. Partnership with Anthem Memory Care In collaboration between Dr. Bodine's Bioengineering Labs at University of Colorado Denver a nd Anthem Memory Care facilities, the research team identified a priority to address currently undetected falls of residents with dementia living at the facility. Despite efforts with current standard of care protocols, there are still falls that go unseen and may or may not be reported after the fact by the resident or by other residents who may have witnessed the fall. The question that this study proposes is: Can a real time, wireless fall detection device detect falls from a standing position and be wor n for 3 days by a resident with dementia in a memory care facility ? Additionally, t he goal is that this device will be used in addition to existing standard of care protocols that Anthem Memory Care has in place to improve quality of care for residents wit h dementia.

PAGE 27

! %H CHAPTER II SPECIFIC AIMS Specific Aim One. Design and build a real time wireless fall detection device using an IMU sensor, microcontroller, and battery. Specific Aim Two. Perform an iterative test using the device on young volunteers betw een the ages of 20 40. These volunteers will mimic eight different falls and complete five daily living activities, which include lying on a bed and then standing, walking a few meters, sitting on a chair then standing, climbing two steps, and standing aft er picking something up. These tests will ensure there are no false alarms when differentiating between falls and daily activities. Specific Aim Three. Build and test a mock wireless fall detection device modeled for the size and weight of the fall detect ion system (FDS). The purpose is to test the mock device on 20 residents at Chelsea Place Anthem Memory Care to demonstrate that the FDS design is unobtrusive to activities of daily life and to ensure that the actual device will not be damaged in the next phase of testing. Specific Aim Four. Test the FDS by integrating the device into a 3 in x 4 inch Band Aid that will be worn by 20 residents at an Anthem memory care facility. The device will operate in addition to the current standard of care. The purpose is to test the FDS with the intended population to validate that the device can detect falls that occur from a standing position by a resident with dementia at Chelsea Place Anthem Memory Care.

PAGE 28

! %$ CHAPTER III MATERIALS AND METHODS Fall Detection System Deve lopment The objective of the fall detection system development was to acquire, a microcontroller, 10 D egree of F reedom I nertial M easurement U nit with a L3GD20H + LSM303 + BMP180 sensor, and battery to be assembled into an initial prototype. A 10 degree of freedom inertial measurement unit is a sensor that has a gyroscope ( L3GD20H), accelerometer/magnetometer combo (LSM303), and a barometric pressure sensor (BMP180). Then a fall detection system was developed This required the development of an algorithm that could collect acceleration and change of altitude from the user's body movements and, using a series of four threshold values, determine if the user had fallen. After this is determined, the algorithm then sends the data samples, which include the va lues from the fall, to a secure broker called a Cloud MQTT A MQTT (Message Queuing Telemetry Transport) is a messaging protocol that provides network clients with a way to distribute and communicate information from machine to machine using a publish/sub s cribe method. The Cloud MQTT then sends the data to a computer The algorithm also connects to the Adafruit MQTT which sends a notification that the user has fallen to a smartphone. The flow of this system is shown in Figure 1.

PAGE 29

! %% Figure 1. Block diagram of the fall detection system. Protocol and Methods The Adafruit Feather HUZZAH ESP8266 WiFi microcontroller was selected because of its small size and ability to connect to the WiFi. The Adafruit 10 Degree of Freedom Inertial Measurement Unit Breakout L3GD20H + LSM303 + BMP180 inertial measurement unit was chosen because it contained an accelerometer, gyroscope, and barometer. These sensors provided 10 degrees of freedom and five data points, yaw, pitch, roll, magnetometer, and barometric readings, to c ollect for fall detection. For the prototype these two pieces were wired together on a breadboard. The 3V, Ground, SCL, SDA pins on the Huzzah Feather ESP8266 were connected to the 3V, Ground, SCL, SDA pins on the Adafruit 10 DOF IMU Breakout respectively, as shown in Figure 2.

PAGE 30

! %& After the IMU sensor and microcontroller were connected in this manner the IMU sensor could send the microcontroller the onboard sensor data. Figure 2. Adafruit Huzzah Feather ESP8266 wired to the Adafruit 10 DOF IMU. A 12 00 mAh 3.7 V Lithium Polymer battery powered the device because it is lightweight and has a long battery life. The Huzzah Feather ESP8266 has a built in battery regulator that only allow s the battery to power the device until it reaches 3.14 V Once it reaches t his voltage the Huzzah Feather ESP8266 automatically shuts off. This results in a battery voltage range of 4.2 V to 3.14 V. Tests to determine battery life were performed using a multimeter to confirm this range Using a voltage divider circuit, seen in Fi gure 3, the battery voltage was stepped down to a max of 0.9 V and a minimum of 0.6 V. The output voltage from the voltage divider was connected to the ADC pin on the Adafruit Feather HUZZAH ESP8266. The voltage was stepped down because the ADC

PAGE 31

! %' pin can onl y read analog voltages between 0 V and 0.9 V. This pin converted the vol tage into a digital range of 680 to 780 Two tests were completed to help determine how battery life is affected by frequency of data deliver. The first test recorded t he output value of the battery every 0 .5 s into a 10 00 row matrix. This matrix was outputted into an excel spreadsheet every 8 min until the battery died. The second test recorded the output value of the battery every 1 s and sent the value directly to an excel spreadshe et immediately after the data point was recorded. After the battery life tests were finished a fall detection algorithm was developed using C++ code on the Arduino platform. (a) (b) Figure 3. (a) Simulation of voltage divider circuit. (b) Voltage reg ulator circuit connected to Fall Detection Device Prototype. The algorithm use s the accelerometer readings, yaw, pitch, and roll, which are identified as rotations around the Z Y and X axes respectively. These measures determine the orientation of the b ody and can be seen in Figure 4. It takes a sample of this output data every 0.25 s. It stores up to 400 rows of data samples in a matrix. When this limit is met it starts re placing the data in the matrix starting with the first row.

PAGE 32

! %( Figure 4. Referen ce system for the Yaw, Pitch and Roll angles ( 4 3). The algorithm examines each sample using a threshold technique in order to look at the impact, aftermath, and posture phases of a fall. The thresholds and definitions for these measures were based on Karat onis' studies ( 46 ), which were also verified by Pierleoni et al. 2015 These values can be seen in Table 1 below. The definitions of these thresholds are as follows: Impact Phase: is the time after the loss of foot contact with the ground and the start o f falling towards the ground, due to attraction of gravity force; the subject impacts on the ground, or other objects that cause an acceleration peak Aftermath Phase: is the immobilization, or nearly so, to the ground associated with low values of accelera tion for a short time Posture Phase: is the end posture, dependent on the direction of the subject's trunk after falling to the ground The algorithm looks at the impact threshold first, then the aftermath threshold, and then finally the posture phase thres hold. An overview of the algorithm is shown in Figure 5.

PAGE 33

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igure 5. Flow chart of the proposed fall detection algorithm. The impact phase threshold is set at 4 g At any point when the root mean square of the acceleration is above that threshold it indic ate s that the impact phase of a fall may have occurred. The root mean square is examined for this threshold because the single combined acceleration value gives a more accurate representation of how the whole body is moving which increases the probability of detecting falls. If this threshold is met the algorithm completes 9 more iterations, at a sampling rate of 0 .01 s, of data collection. Each of these data samples is examined to determine if they meet the next threshold. If they do not, the code resume s sampling data every 0.25 s and looking for a spike in the root mean square value.

PAGE 34

! %* The next threshold that is examined is the aftermath phase. It is detected when the root mean square of acceleration presents an almost flat trend close to 1 g The algor ithm checks the lower and upper bounds, 1 g and 2 g respectively, of the threshold for 1 s following the impact threshold being reached It also checks to see if there was an altitude change of 0.4 m. If this threshold was met the algorithm moves onto th e last criteria. If it was not met the code finishes sampling and examining the rest of the 9 data points. Once finished it then moves back to sampling data every 0.25 s and looking for a spike in the root mean square value. The last threshold is the p osture phase, which is defined by orientation, which is an angle within a specified range within 1 s after the aftermath phase. The posture phase is observed by looking for changes in p itch and r oll. The p itch and r oll values change based on whether the p erson has fallen right, left, backwards, or forwards. The specific values for each direction are listed in Table 1. If this final threshold is met the code assumes that a fall has occurred At this point, the algorithm then connects to WiFi and connects t o the Cloud MQTT and Adafruit MQTT A n additional 100 data points are then recorded at a sampling rate of 0.001 s, to ensure that the data from the entire fall was recorded. Table 1 Shows the phases of a fall, the important parameters of that phase, and the corresponding threshold values that are being used. Phase of Fall Parameter of Interest Threshold Value Impact Phase Mean Acceleration x> 4 g Aftermath Phase Mean Acceleration 1 g
PAGE 35

! %+ I t then takes the 400 data points that are populating the matrix and packages it into a JSON packet. JSON (JavaScript Object Notation) is a standard format to send data objects in an attribute value pair. One row of the JSON packet contains : RMS: Corresponding Value, Altitude: Corresponding Value, Roll: Corresponding Value, Pitch: Corresponding Value, an d Time: Corresponding Value. This is repeated for each of the 400 rows. This packet is then encrypted and securely published to the Cloud MQTT under the name Fall 2. It was encrypted and securely published using a Secure Sockets Layer port between the Cl oud MQTT and its connected clients. It is then sent to any client that is connected to Cloud MQTT and subscribed to accept any message or data with the name Fall 2. In this case the client is a computer that is connected to Cloud MQTT using a python progr am and is subscribed to receive any messages that are sent to the Cloud MQTT server with the title Fall 2. The Cloud MQTT takes the message from the fall detection sensor and then sends it to the computer. This entire process takes about 0 .18 s The pyth on code then takes the JSON packet and examines it one row at a time. It separates the values in the row based on the labels. It places the separated pieces of data into the corresponding heading within the comma separated values (CSV) file titled test.csv The fall detection device als o sends the number 1 to the Adafruit MQTT feed f all 3. This changes the feed fall3 from 0 to 1. This triggers the application IFTT, which was monitoring the feed fall3, to send the notification Fall has o ccurred to the smartp hone as seen in Figure 6 IFTT stands for If This Then That and is a web based service that allows users to make conditional statements called Applets. The Applets work by allowing the developer to dictate if this happens then that happens. In this case

PAGE 36

! %G if Adafruit feed fall3 equals 1 then a notification message "Fall has occurred" is sent to the smartphone with the IFTT application. It can take 2 to 5 minutes for this notification to occur. Figure 6. Fall detection notification shown on a smartphone Figure 7 shows how the Fall Detection Device, Computer, and IFTT Application work together with the Cloud MQTT and Adafruit MQTT

PAGE 37

! &H Figure 7. Block diagram showing how the client/broker system works. Young Healthy Volunteer Study The objective of the youn g healthy volunteer study was to determine if the fall detection system could accurately and consistently detect 9 different types of falls and not classify four different activities of daily living as falls. Sites The a ssistive technology partner's lab w as used to test the fall detection system on the young healthy volunteers. The set up for the study which can be seen below, consisted of two teal pillows, and two couch cushions covered by the yellow and blue blanket. These are on top of two gym mats com monly used in gymnastics.

PAGE 38

! &$ Figure 8 Set up of the test site at assistive technology partners lab. Subjects The study was completed using 18 healthy individuals between the ages of 20 56 who have worked with people with disability. Table 2 shows age, gen der, and height Table 2. Demographic information about the subjects. Gender Height Age Female 5'7 23 Female 5'3 24 Female 5'0 24 Female 5'8 23 Female 5'6 25 Male 5'9 25 Female 5'4 24 Female 5'9 23 Male 6'1 24 Female 5'4 23 Male 5'6 23 M ale 6'4 26 Male 6'1 26 Female 5'5 55 Female 5'7 56 Female 5'9 28 Female 5'10 30 Male 6'0 55

PAGE 39

! &% Consenting Study included people testing the prototype by completing falls and being video taped with a camera. Subjects who were video taped at ATP con sented by signing the Colorado Multiple Institutional Review Board approved consent forms under the study, "An Exploratory Investigation of the Impact of the Assistive Technology Partners Product Testing Lab." These forms included a consent packet and a ph otograph/videotape release form. The form were read by each subject and reviewed verbally in a quiet section of ATP before signing. Protocol and Methods Following the Usability Protocol (COMIRB #11 067), the study was completed using 18 individuals betwe en the ages of 20 56 who have worked with people with disability. Participants wore a strap around their chest and the FDS was securely placed underneath the strap for the duration of the study, which is shown in Figure 9 below Figure 9 The blue s quare shows how the fall detection device was placed under the strap that was secured around the participant's chest.

PAGE 40

! && The participants completed 9 different fall types into a soft padded mattress. The types of falls are listed below: Fall backwards and en d on left or right shoulder Fall backwards and end on back Fall forward and end on stomach Lateral left fall and end on stomach Lateral left fall Lateral right fall and end on stomach Lateral right fall Fall backwards, land on butt, and end lying on back They were then asked to complete four activities of daily living, which are listed below: Walk 5 meters Sit down and stand back up Bend down to pick up a pencil from the floor and stand back up Lay down and then stand back up Data Collection The componen ts in the inertial measurement unit recorded the subject's body movements. The accelerometer recorded the subject's acceleration in the X Y and Z axes. T he gyroscope and magnetometer readings were recorded to help determine if they were sitting, standing upright, or lying down, and the barometric readings were recorded to determine if they are still standing or not. When each of these components reached a specific threshold a fall was detected Once a fall was detected 400 data points, 180 before the fall and 20 after the fall, were se n t to a computer excel spreadsheet

PAGE 41

! &' Data Analysis Once each subject had completed the protocol, the number of falls detected by the device, the number of falls that were no t detected but were witnessed by the principle inves tigator and the number of false alarms the device detected were tallied False alarms are defined as the device send ing a fall alert when the user did not fall. Then, sensitivity, specificity and accuracy were calculated as described below: 1) Sensitivity: C apacity to detect a fall, which is calculated as the ratio of true positives and the sum of true positives and false negatives 2) Specificity: Ability to avoid detection of a normal event as a fall, which is calculated as the ratio of true negatives and the s um of true negatives and false positives 3) Accuracy: Capacity of correctly detecting a fall, which is calculated as the ratio of true positives and the sum of true positives and false positives Additionally, t he accuracy average and standard deviation of the fall direction data were calculated. V isual graph s of each fall direction were also created to show the difference between each direction. The falls that were not detected were examined in order to determine why that was the case Mock Device Study T he mock device phase of the study was completed to determine whether the device could be integrated into the daily routine of residents who live in a memory care facility without ca using irritation It also test ed whether or not the FDS, which is more cost ly, would be destroyed when it was worn for testing.

PAGE 42

! &( Sites The sites used for this study were Chelsea Place Anthem Memory Care in Aurora. Subjects 10 residents at Chelsea Place Anthem Memory Care were recruited to wear the mock device in addition to continuing to receive standard of care. 9 of them agreed to participate and their demographics are listed in Table 3. Inclusion and exclusion criteria for this population are listed below. Inclusion Diagnosed with dementia Willingness and capability to giv e informed consent to participate in the study, or consent of a legally appointed guardian if he/she is unable to provide full consent according to institutional guidelines Primary language is English Living at Chelsea Place Anthem Memory Care Is able to w alk around Chelsea Place Anthem Memory Care on their own Exclusion Residents at Chelsea Place Anthem Memory Care facility that primarily use a wheelchair Residents at Chelsea Place Anthem Memory Care facility that are on end of life hospice care Residents at Chelsea Place Anthem Memory Care facility that have a history of developing skin rashes from Band Aids

PAGE 43

! &) Residents at Chelsea Place Anthem Memory care who have a current injury that prevents mobility Table 3 R esident demographics; gender and ambulatory s tatus !"#$%&'( ) ( !%* ( +,% ( -./,012.2 ( 3"44%0'( +5#"6/'147(!'/'"2 ( 8 ( 9/6% ( :; ( +6<=%.5%4 >2 ( ?/6@2(A.'=(A/6@%4( .B(=%(4%5%5#%42 ( C ( 9/6% ( DE ( +6<=%.5%4 >2 ( ?/6@2(10(=.2(1A0 ( F ( G%5/6% ( DH ( +6<=%.5%4 >2 ( ?/6@2(10(=%4(1A0 ( ; ( G%5/6% ( EI ( +6<=%.5%4 >2 ( 3/0%JA/6@%4JA/6@2( 10(=%4(1A0 ( I ( G%5/6% ( EH ( +6<=%.5%4 >2 ( ?/6@%4 ( K ( 9/6% ( :: ( +6<=%.5%4 >2 ( 3/0% ( D ( G%5/6% ( :8 ( L/2&"6/4(-%5%0'./ ( ?/6@2(10(=%4(1A0 ( : ( G%5/6% ( :D ( +6<=%.5%4 >2 ( ?/6@2(10(=%4(1A0 ( E ( G%5/6% ( DI ( +6<=%.5%4 >2 ( ?/6@2(10(=%4(1A0 ( Consenting After the legally authorized representative (LAR) v olunteered, the LAR was asked to meet with the trained study staff/investigator to read, ask questions, and sign the consent form. This meeting occurred in a quiet, private meeting room at Chelsea Place Memory Care facility. The room was outfitted with a t able, chairs, and any necessary accommodations needed by the potential subject to comfortably and fully participate in the consenting process (e.g. large print documents for someone with a visual impairment) with no external distractions. The trained stud y staff/investigator read through the consent form with the LAR. The LAR was asked if they had any questions or concerns, and was informed and assured that

PAGE 44

! &* they could remove their loved one from the study at any time and no one would be mad. Once the ir que stions had been answered, the trained study staff/investigator left the room and allowed the LAR to think about whether they wanted their loved one to participate in the study. When a decision was reached, the LAR opened the door to the room to signify tha t they had reached a decision and the trained study staff/investigator came back into the room to hear the decision. If the LAR decided to participate, all required parties signed the consent form. Protocol and Methods 9 s ubjects wore the mock device f or 3 days. The mock device was the same size and weight as the actual fall detection system. The total weight was approximately 35 grams. It was integrated into a Band Aid intended to avoid irritating the skin and was similar to existing patches that some residents wear for pain relief. It was worn by residents on their mid back or lower shoulder (see F igure 10 ) The trained carestaff at Chelsea Place Memory Care placed the Band Aid and monitored t hem daily. If the resident became irritated or agitated by the device or developed a skin irritation from the device, the carestaff were instructed to immediately remove it Figure 10 Mock device that was placed on the mid back or lower shoulder of subjects.

PAGE 45

! &+ Data Collection The carestaff monitored the mo ck device on the resident throughout the day. If a user was irritated or developed a skin irritation from the mock device it was reported to the PI and they were removed from the study. If the mock device fell off overnight, while walking around, during an organized activity, or while sitting it was reported to the PI. Final Device Testing The final device testing ensured that the system could be integrated into the daily routine of residents at a memory care facility without it being aggravating or irrita ting while also detecting any falls that occur while they wore the device. Sites The sites used for this study were Chelsea Place Anthem Memory Care in Aurora. Subjects These subjects are the same subjects as the mock device study, excluding anyone th at became irritated by device or developed a skin rash from it Inclusion and exclusion criteria for this population are listed below. Inclusion Diagnosed with dementia Willingness and capability to give informed consent to participate in the study, or co nsent of a legally appointed guardian if he/she is unable to provide full consent according to institutional guidelines Primary language is English Living at Chelsea Place Anthem Memory Care

PAGE 46

! &G Is able to walk around Chelsea Place Anthem Memory Care on their own Exclusion Residents at Chelsea Place Anthem Memory Care facility that primarily use a wheelchair Residents at Chelsea Place Anthem Memory Care facility that are on end of life hospice care Residents at Chelsea Place Anthem Memory Care facility that hav e a history of developing skin rashes from Band Aids Residents at Chelsea Place Anthem Memory care who have a current injury that prevents mobility Table 4 R esident demographics; gender and ambulatory status !"#$%&'() ( !%* ( 3"44%0'(+5#"6/'147( !'/'"2 ( 8 ( 9/6 % ( ?/6@2(A.'=(A/6@%4(.B(=%( 4%5%5#%42 ( C ( 9/6% ( ?/6@2(10(=.2(1A0 ( F ( G%5/6% ( ?/6@2(10(=%4(1A0 ( ; ( G%5/6% ( 3/0%JA/6@%4JA/6@2(10( =%4(1A0 ( I ( G%5/6% ( ?/6@%4 ( K ( 9/6% ( 3/0% ( D ( G%5/6% ( ?/6@2(10(=%4(1A0 ( : ( G%5/6% ( ?/6@2(10(=%4(1A0 ( E ( G%5/6% ( ?/6@2(10(=%4(1A0 ( Additionally, al l carestaff that work ed with the subjects in this study were asked to complete a consent form and take a short usability survey.

PAGE 47

! 'H Consenting The consenting process completed in the mock device study carrie d over to the final device testing. The caret akers that volunteered to take the survey were consented after the final device testing was complete d The consent meeting for any caretakers who volunteered to take the survey took place in a quite private meeting room at Chelsea Place Memory Care facilit y. The trained study staff/investigator read through the postcard consent form with the caretaker. The caretaker was asked if they have any questions or concerns, and was informed and assured that they could choose not to fill out or finish the survey at a ny time and no one would be mad. Once the questions had been answered, the trained study staff/investigator left the room and allowed the caretakers to think about if they are willing to participate. When a decision had been reached, the participant opened the door to the room to signify they had reached a decision and the trained study staff/investigator came back into the room to hear the decision. If the caretaker decided to participate, they filled out the survey following the postcard consent form. P rotocol and Methods The final phase of this protocol included having 9 residents wear the FDS seen in Figure 11, for 3 days. The total weight of the FDS was approximately 35 g. It was integrated into a Band Aid intended to avoid irritating the skin and was similar to existing patches that some residents wear for pain relief. It was worn by the residents on their mid back or lower shoulder. The trained staff at Chelsea Place Memory Care placed the Band Aid and monitored it daily. If the resident became i rritated or agitated by the device or developed any skin irritation, the carestaff immediately removed the device.

PAGE 48

! '$ Figure 11 Final Fall Detection Device that was placed on the mid back or lower should of the subjects. Data Collection The components in the inertial measurement unit recorded the subject's body movements. The accelerometer recorded the subject's acceleration in the X Y and Z axes. T he gyroscope and magnetometer readings were recorded to help determine if they were sitting, standing up right, or lying down, and the barometric readings were recorded to determine if they are still standing or not. When each of these components reached a specific threshold, a fall was detected T he fall data was then sent to a computer, an alert that the re sident had fallen was triggered, and a notification was sent to the caretakers Anthem Memory Care standard issued iPod. Facility staff then implemented their Standard of Care Fall routine which is outlined below: First your ward will be evaluated to dete rmine if severe pain exists and if he/she can move. If he/she has severe pain they should not be moved, and the emergency services will be summoned. If an Anthem Memory Care nurse is in the community, an evaluation and assessment may be

PAGE 49

! '% completed prior to notifying EMS. If after the licensed nursing assessment is complete, the nurse may direct to NOT notify EMS. The Primary care physician must be notified. Next, the Anthem Memory Care staff will determine if your ward is weight bearing and can assist staff members in getting them up from the floor. Then if they feel like they can bear weight, two employees can provide standby assistance to help the resident get up from the floor. If possible, staff members should provide reassurance and calmly talk to them throughout the process. If they cannot assist and is unable to bear weight, the care staff will make them as comfortable as possible and call emergency services. They will offer pillows for support, and blankets for comfort. The care staff will stay with y our ward at all times until emergency services arrive. Next an Incident Report form shall be fully completed and placed in the Incident Report binder. Repeated falls by your ward will have follow ups to include a Fall Risk Assessment and progressive measu res to help avoid future falls. If your ward falls and strikes their head, there is a risk for a closed head injury and, therefore, 911 is to be called and evaluate the need to be sent to the ER or seen by a MD for evaluation of the injury.

PAGE 50

! '& A Head Injury M onitoring Form is to be implemented for all falls with suspected or actual head injury involvement. The daily reports about how the subject tolerated the device, and how well it stayed on the subject were also recorded. Additionally, the system usability surveys that the caretakers filled out were collected. Data Analysis Once each subject had c ompleted the protocol, the number of falls detected by the device, the number of falls that were not detected but were witnessed by care staff in the facility, and the number of false alarms the device detected were tallied False alarms are defined as the device send ing a fall alert to the caretaker when the user did not fall. Then, sensitivity, specificity and accuracy was calculated as described below: 4) Sensitivit y: Capacity to detect a fall, which is calculated as the ratio of true positives and the sum of true positives and false negatives 5) Specificity: Ability to avoid detection of a normal event as a fall, which is calculated as the ratio of true negatives and t he sum of true negatives and false positives 6) Accuracy: Capacity of correctly detecting a fall, which is calculated as the ratio of true positives and the sum of true positives and false positives Once this was completed, the scores from the system usabili ty survey were tallied in order to determine what the caretakers thought of the device.

PAGE 51

! '' CHAPTER IV RESULTS Fall Detection System Development The battery life of the 3.7 V 12 00 mAh lithium polymer battery while running the fall detection algorithm witho ut sending any data points was 36 hrs. The battery life while sending 1000 data points once every 8 min was 2 0 h rs The discharge of the battery, seen in the figure below, was what is expected of a lithium polymer battery. !" !#$" %" %#$" &" &#$" '" '#$" (" (#$" !" $" %!" %$" &!" &$" !"#$%&'()!*( +,-'()./0*( Figure 12. B attery voltage re ading throughout the 2 0 hrs of battery life. The battery life while sending the data point immediately after recording the point was 12 hrs. The accelerated discharge of the battery can be seen in the figure below.

PAGE 52

! '( !" !#$" %" %#$" &" &#$" '" '#$" (" (#$" !" &" (" )" *" %!" %&" !"#$%&'()!*( +,-'()./0*( Figure 13. Battery voltage reading duri ng the 12 hrs of battery life. It was also found that it takes .18 seconds to send the data from th e fall detection device to the computer. Additionally, it can take 2 to 5 minutes for the fall detection device to send a notification to the smartphones thr ough the IFTT phone application. Young Healthy Volunteer Study The 18 subjects completed 9 different falls and 4 activities of daily living (ADL). In total 164 falls and 72 ADLs were completed. The fall detection system did not identify any of the ADLs as a fall. The number of true positives was 155, where a true positive is defined as a fall that was correctly detected. The number of true negatives was 72, where a true negative is a n activity of daily living that is not misidentified as a fall. There we re 0 false positives, where a false positive is defined as a non fall that is misidentified as a fall. There were 9 false negatives, where a false negative is a fall that was not detected. Of the 9 falls not detected, 5 of them occurred when the subject fe ll to the right. Table 5 shows the results of each fall for each of the 18 subjects. Table 6 show s the f all directions attempted by each subject and the number of falls detected and not detected

PAGE 53

! ') Table 5. Experimental results from protocol: Fall Detected(x), Fall Detec ted on Second Fall X(2), Fall Not Detected(Blank Box), Fall Not Completed ( ), Correct Fall Indications in Black, Incorrect Fall Indications in Red

PAGE 54

! '* Table 5 cont. Experimental results from protocol: Fall Detected(x), Fall Detected on Second Fall X(2), Fal l Not Detected(Blank Box), Fall Not Completed ( ), Correct Fall Indications in Black, Incorrect Fall Indications in Red

PAGE 55

! '+ Table 5 cont. Experimental results from protocol: Fall Detected(x), Fall Detected on Second Fall X(2), Fall Not Detected(Blank Box), F all Not Completed ( ), Correct Fall Indications in Black, Incorrect Fall Indications in Red

PAGE 56

! 'G Table 6. Fall directions attempted by each subject and the number of falls detected and not detected. Fall Direction Total # Falls Attempted # Falls Detected # Falls Not Detected Fall backwards end on left/right shoulder 18 17 1 Fall backward s 19 18 1 Fall forward end on left/right shoulder 18 18 0 Fall forward 18 18 0 Lateral fall left end on stomach 17 17 0 Lateral fall left 18 17 1 Lateral fall right end forward 17 17 0 Lateral fall right 20 15 5 Fall backwards, land on butt, end on back 19 18 1 The results from the subjects had a sensitivity, specificity, and accuracy of 95%, 100%, and 96% respectively seen in Table 7 Table 7. Number of falls correctly identified (True Positives), Activities of Daily Living not detected as fall (True Negative), Number of Activities of Daily Living detected as falls (False Positives), Number of falls not detected as falls (False Negatives), and the sensitivity, specificity, and accuracy. True Positives 155 True Negatives 72 False Positives 0 False Negatives 9 Sensitivity 0. 9 5 Specificity 1 Accuracy 0.96

PAGE 57

! (H The reason the fall detection device did not detect 8 of the 9 falls was due to either the p itch or r oll value being outsi de of the threshold values seen in the table below. The lateral fall right completed by Subject 3 was not detected despite all values being above the threshold values. The reason for this may have been because only one set of pitch and roll values were above the threshold and the algorithm may have missed this. Table 8. The nine falls that were not detected and a description of why it was not detected. Subject and Fall Direction Reason Fall Was Not Triggered Subject 3 Lateral Fall Right All Values Above Their Respective Thresholds Subject 5 Lateral Fall Right Pitch Pitch Threshold Subject 10 Fall Backwards, Land on Butt, E nd on Back Roll Pitch Threshold Subject 9 Fall Right Pitch>Pitch Threshold Subject 10 Fall Backw ards RollPitch Threshold Subject 11 Fall Backwards End Left PitchPitch Threshold Subject 15 Lateral Fall Left Pitch
PAGE 58

! ($ Table 9. The fall direction and the number of directions correctly identified for each fall type. Fall Direction # Falls Direction Correctly Identified Fall B ackwards E nd L eft/ R ight 14 Fall B ackwards 16 Fall F orward E nd L eft/ R ight 14 Fall F orward 8 Lateral F all L eft E nd F orward 2 Lateral F all L eft 10 Lateral F all R ight E nd F orward 2 Lateral F all R ight 12 Fall Backwards, Land on Butt, E nd on Back 15 Table 10. The number of fall directions that were misidentified for each fall type and the direction that was indicated for each misidentification. T he algorithm was written so that it could detect whether the person had fallen and whether they ended laterally to the right, laterally to the left, on their back, or on their Fall Direction the Se nsor was Supposed to Indicate # Of Fall Directions Misidentified Fall Direction the Sensor Indicated for Each Misidentification Fall B ackwards E nd L eft 2 2 Right Fall B ackwards E nd Right 0 NA Fall B ackwards 0 NA Fall F orward E nd L eft 2 1 Left and 1 Forward Fall Forward End Right 0 NA Fall F orward 9 9 Right Lateral Fa ll L eft E nd F orward 12 12 Right Lateral F all L eft 5 5 Right Lateral F all R ight E nd F orward 14 14 Right Lateral F all R ight 2 1 Backward and 1 Left Fall Backwards, Land on Butt, E nd on Back 3 3 Right

PAGE 59

! (% stomach. It was split in these four directions be cause each direction had a unique range of RMS, Altitude, Pitch, and Roll values. Creating different threshold ranges for RMS, Altitude, Pitch, and Roll for each direction leveraged this This can be seen in the distinctive graph patterns for each fall dir ection (Figures 14 17) The visual representation from a subject ending on a left shoulder is seen below in Figure 14 The four graphs show the output of the Roll (Figure 14a), Pitch( Figure 14b), Altitude (Figure 14c), and RMS (Figure 14d) over the period before the fall, the impact, aftermath and posture phases. These are pointed out in the figure. When a subject ends on the left shoulder the RMS spikes, the altitude decreases, the roll value increases between 32 deg and 15 deg and the pitch value decre ases between 149 deg and 71 deg This pattern is reflected throughout the rest of the study sample. 840 850 860 870 Time (ms) -80 -60 -40 -20 0 20 Degrees (¡) Subplot Roll 840 850 860 870 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 840 850 860 870 Time (ms) 1586 1588 1590 1592 1594 1596 Estimated Altitude (m) Subplot Altitude 840 850 860 870 Time (ms) 0 5 10 15 20 Acceleration (g) Subplot RMS Altitude Change Impact Aftermath Posture Posture !"#$ !%#$ !&#$ !'#$ Figure 14 Example roll, pitch, altitude, and RMS data from a fall onto the left shoulder.

PAGE 60

! (& The visual representation of the data from a subject ending on a right shoulder is shown in Figure 15 Figure 15 show s the output of the Roll (Figure 15a), Pitch (Figure 15b), Altitude (Figure 15c), and RMS (Figure 15d) over the period before the fall, the impact, aftermath, and posture phases. These are pointed ou t in the figure. When an individual ends on the right shoulder the RMS value spikes, the altitude decreases, the roll value increases between 25 deg and 17 deg and the pitch value decreases 27 deg and 1 2 0 deg This pattern is reflected throughout the en tire sample 210 220 230 240 250 Time (ms) -100 -50 0 50 Degrees (¡) Subplot Roll 210 220 230 240 250 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 210 220 230 240 250 Time (ms) 1578 1580 1582 1584 1586 Estimated Altitude (m) Subplot Altitude 210 220 230 240 250 Time (ms) 0 2 4 6 8 10 12 Acceleration (g) Subplot RMS Posture Posture Altitude Change Impact Aftermath !"#$ !%#$ !&#$ !'#$ Figure 1 5 Example roll, pitch, altitude, and RMS data from a fall onto the right shoulder. The visual representation from a subject ending on the back is seen below in Figure 16 Figure 16 shows the output of the Roll (Figure 16a), Pitch (Figure 16b), Altitude (Figure 16c), and RMS (Figure 16d) over the period before the fall, the impact,

PAGE 61

! (' aftermath, and posture phases. These are pointed out in the figure. When a subject ends on their back the RMS spikes, altitude decreases, the roll value increases between 10 deg and 7 deg and the pitch value decreases between 230 deg and 18 deg This pattern is reflected throughout the rest of the study sample. 325 330 335 340 345 350 Time (ms) -100 -50 0 50 Degrees (¡) Subplot Roll 325 330 335 340 345 350 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 325 330 335 340 345 350 Time (ms) 1565 1570 1575 1580 1585 1590 1595 Estimated Altitude (m) Subplot Altitude 325 330 335 340 345 350 Time (ms) 0 2 4 6 8 Acceleration (g) Subplot RMS Posture Posture Impact Aftermath Altitude Change !"#$ !%#$ !&#$ !'#$ Figure 1 6 Example roll, pitch, altitude, and RMS data from a fall onto the back. The vi sual representation from a subject ending on the stomach is seen below in Figure 17 Figure 17 show s the output of the Roll (Figure 17a), Pitch (Figure 17b), Altitude (Figure 17c), and RMS (Figure 17d) over the period before the fall, the impact, aftermath and posture phases. These are pointed out in the figure. When a subject ends on their forward the RMS spikes, altitude decreases, the roll value increases between 26 deg and 3 deg and the pitch value decreases between 10 deg and 39 deg This pattern i s reflected throughout the rest of the study sample.

PAGE 62

! (( 70 75 80 85 90 Time (ms) -80 -60 -40 -20 0 Degrees (¡) Subplot Roll 70 75 80 85 90 Time (ms) -200 -150 -100 -50 0 50 Degrees (¡) Subplot Pitch 70 75 80 85 90 Time (ms) 1578 1580 1582 1584 1586 Estimated Altitude (m) Subplot Altitude 70 75 80 85 90 Time (ms) 0 5 10 15 20 Acceleration (g) Subplot RMS Posture Posture Impact Aftermath Altitude Change !"#$ !%#$ !&#$ !'#$ Figure 1 7 Example roll, pitch, altitude, and RMS data from a fall onto the stomach. The pitch and roll values that were recorded after the RMS threshold was triggered for detected falls and falls that went undetected were collected They were then sorted by fall direction in order to determine the range of pitch and roll values that occurred during testing The averaged pitch and roll values for falls w h ere the subject ended on their left shoulder are shown in Figure 1 8 Each fall that ended with the subject on their left shoulder had an average roll value that was extremely close to one another. The average pitch value was close for two of the falls and within one standard deviation of the third.

PAGE 63

! () Table 11. Resulting average, standard deviation, and range of pitch and roll values for each type of fall

PAGE 64

! (* ! ! ! ! ! ! ! Figure 1 8 Average pitch and roll values for falls that ended with the subject on their left shoulder with standard deviation error bars The average pitch and roll values for falls that ended with the subject on thei r right shoulder were all within 20 deg of each other ( Figure 1 9 ) The average roll values were once again closer together than the average pitch values. ! ! Fall Backwards End on Left Shoulder Fall Forward End on Left Shoulder Lateral Left Fall

PAGE 65

! (+ ! ! ! ! ! ! Figure 1 9 Average pitch and roll values for falls that ended with the subject on their right shoulder with standard deviation error bars The average pitch and roll values for falls that ended with the subject on their stomach were both within 20 deg for each fall (Figure 20) The average roll value was within 6 deg of each other whereas the average pitch values were within 11 deg. ! ! Fall Forward End on Right Shoulder Fall Backwards End on Right Shoulder Lateral Right Fall

PAGE 66

! (G ! ! ! ! ! ! Figure 20 Average pitch and roll values for falls that ended with the subject on their stomach with standard deviation error bars The a verage pitch and roll values for falls where the subject ended on their back had a close cluster for both values (Figure 21) ! ! ! Lateral Fall Right End on Stomach Fall onto Stomach Lateral Fall L eft End on Stomach

PAGE 67

! )H ! ! ! ! ! ! ! Figure 21 Average pitch and roll values for falls that ended with the subject on t heir back with standard deviation error bars !! Mock Device Study Of the 9 subjects that participated in the study all 8 were able to wear the mock device for the full 3 days. One of the subjects did refuse to have the mock device placed on them the fi rst morning due to multiple irritations and frustrations that they were experiencing. When the subject was re approached later in the day after they had calmed down, they agreed to wear it During the three days the subjects wore the mock device it stayed on all 3 nights for 1 subject and fell off 1 out of the 3 nights for 3 subjects It fell off 2 out of the 3 nights for 3 subjects and 3 out of 3 nights for 0 subjects. Only 1 Fall Backwards, Land in Sitting Position, End on Back Fall Backwards

PAGE 68

! )$ subject took the mock device off while wearing it. Table 12 show s the days tha t the mock device stayed on the times the residents took it off, and the times it fell off at night for each subject The NA entries in the table were because the carestaff did not have time to place the device on the subject that day.

PAGE 69

! )% Table 12. Total number of times each subject had the mock device fall off at night, took the mock device off during the day, refused to wear it, and wore it throug hout the night.

PAGE 70

! )& Final Device Te sting Of the 9 subjects that participated 7 were able to we ar the FDS for the full 3 days. During the three days the 7 subjects wore the FDS no one fell while wearing it. Additionally, none of the activities of daily living were detected as falls whil e the subjects wore the FDS. Subject 5 was not approached on the first day to wear the device due to the irritations and frustrations they were having with the day and then on the second and third day they said they did not want to wear it Subject 8 took the device off twice during the first day due to irritation. The subjected stated that it was itchy and bothering her. Due to this reaction they were not asked to wear the device again during day two or three. While wearing the real device, it did not fal l off at all for 2 subjects and it fell off 1 out of the 3 nights for 1 subject. It fell off 2 out of the 3 nights for 3 subjects and 3 out of 3 nights for 0 subjects. It fell off during the day for only one subject. Table 13, shown below, shows the days that the mock device stayed the times the residents took it off, and the times it fell off at night for each subject. Table 14, shown below, shows the results from the system usability survey. Overall, the average score from the Chelsea Place Memory Care staff was a 57 /100 The NA entries in the table were because the carestaff did not have time to place the device on the subject that day

PAGE 71

! )' Table 13. Total number of times each subject had the real fall device fall off at night, took the real fall device off during the day, refused to wear it, and wore it throughout the night.

PAGE 72

! )( Table 14. Chelsea Place Memory Care staff system usability survey results. !"#$%&'() ( !72'%5(M2/#.6.'7(!&14% ( 8 ( KDNI ( C ( ;CNI ( F ( II ( ; ( IH ( I ( DCNK ( +O%4/,% ( IDNIC (

PAGE 73

! )) CHAPTER IV DISCUSSION AND FUTURE WORK Fall Detection System Development The fall detection algorithm was split into three thresholds b ased on the success of this method in Pierloni et al., 2015's resear ch This method was successful because each parameter change s when a person falls and using more than one of those parameters improves the likelihood o f accurately detecting a fall. T he final threshold, pitch and roll was broken down into four different directions to help improve the identification of how a person landed and where an injur y may be. The range of values for each direction were based on the range of values the sensor output ted when it was resting in each of the four directions. The battery life of the system was also an important consideration when designing the device. Multiple steps to improve the battery life while keeping the design lightweight and compact were taken. For example the data sampling rate was decreased to 0 .25 s, and the FDS only connected to the WiFi and MQTT after all four thresholds were met. These changes nearly doubled the overall battery life of the FDS Two different lithium polymer batteries that could power the device for more than a day were identified One was a smaller battery (23 g, 34mm x 62mm x 5mm) that would last for a little over a day and the other was a larger battery (52g, 51mm x 65 mm x 8mm) that would last for 3 days. The caretakers at Chelsea Place Memory Care were asked which one they would prefer. They stated that they wanted to check on the device every morning during testing, thus they preferred the smaller battery that lasted for a day.

PAGE 74

! )* The smartphone application IFTT was the last step in the fall detection system. The fall detection system tak es 2 to 5 minutes to send the fall notification. The reason for this is because there is a delay in how often it checks the fall3 feed on the Adafruit MQTT. In the future, a smartphone application needs to be developed specifically for this system so that the notifications can be sent quickly and efficiently Young Healthy Volunteer Study This study help ed validate the algorithm used for the Fall Detection System. The results of 95% sensitivity, 100% specificity, and 96% accuracy, while not as high of sensi tivity and accuracy as Pierloni et al., 2015, validated that the fall detection system is able to detect a large number of falls by people of varying height and age. The 100% specificity is critical because it proves that the device would not send a false alarm to an already very busy caretaker. Of the nine falls that were not detected 5 of them were lateral falls to the right, two were falls that ended on the back, and one was a fall backwards end ing on the left shoulder. Each of the falls that were not d etected met the RMS, aftermath, and altitude threshold. The reason 8 of the 9 falls were not detected was because the pitch and roll values were outside of the threshold. The last fall that was not triggered met all of the thresholds, but did not send a fi nal notification. This may of happened because all four thresholds were only met for one data point, which could have caused the algorithm to miss the fall indication. The algorithm can be improved to detect an even higher percentage of falls by altering t he threshold ranges to reflect the range of values that indicated a fall in the test data The directional indication of the fall detection system was 66% accurate. The low percentage value can be attributed to the large number of falls that were classifi ed as falls

PAGE 75

! )+ ending on the right shoulder. Of the 43 falls that were misidentified 39 of them were misidentified as lateral right fall One reason for this is that the pitch and roll threshold for lateral right fall had the largest range This large range resulted in significant overlap with the pitch and roll threshold ranges for the other three fall directions. Another reason for the low accuracy percentage is that when the subjects fell o nto the matt they would land in the correct orientation, bounce in to a different orientation, and then stand up quickly The multiple movements immediately after the fall may have affected the algorithms ability to determine the correct orientation. Further examination of the values from the recorded falls could provide more accurate values for the pitch and roll thresholds for future testing. The unique pitch and roll values for each fall direction resulted in distinctive graph patterns The graphs also show the three phases of a fall : impact, aftermath and posture, and altitude change. The altitude change is consistently between 0 .5 and 2 m. The impact value while easily visible by the sharp RMS increase, changes based on how fast the person fell T he aftermath phase that follows is characterized by the flat accelerati on line around 1g The values of pitch and roll values are the only values that change based on the direction of the f all The pitch and roll values that were recorded after the RMS threshold was triggered for detected falls and falls that went undetecte d were collected. They were then sorted by fall direction in order to determine the range of pitch and roll values that occurred during testing These values were averaged and a standard deviation was found. The standard deviation was found in order to d etermine more accurate fall direction threshold ranges. The resulting average values for the falls that ended on the left shoulder

PAGE 76

! )G were all within a standard deviation of each other. The same can be said for falls that ended on the right shoulder, stomach, and back. This shows that falls ending in the same position resulted in similar pitch and roll values. These averaged values and standard deviations provide new more accurate pitch and roll ranges for each of the four fall directions. Mock Device and F inal Device Study The mock device stayed on for approximately half of the nights it was worn and fell off for the other half The real device stayed on for a little under half of the nights it was worn. One subject wore the mock device for two days. On the third day they took off the mock device, and then when they transitioned to wearing the real device they took that one off as well. The subject stated that it was itching her Due to this irritation she was not asked to wear it any longer A second sub ject also took the real device off on the third day they were wearing it. These results suggest that this placement is a good location for most subjects, and that it will not bother them. It also shows that for some individuals an addition of a Band Aid wi ll cause too much irritation. Creating a thinner device with a smaller battery that can fit within the Band Aid could solve the irritation that some subjects felt. It could also solve the issue of the device falling off every other night. The device only fell off one subject during the day because the adhesive would no longer stick to the person. This could have been because the subject only bathes once every three to four days and their skin was too oily. Further research into better adhesive quality sho uld be done. Additionally, one of the sensors was lost after a resident took off the device when a caretaker was not around. This device was found a day later after the

PAGE 77

! *H subjects room was cleaned T his is a common occurrence with many objects in a memory ca re facility and needs to be taken into consideration when designing a device for this population in the future One subject either refused to wear the device or were not approached to wear the device due to the mood that they were in three days in a row These results highlight the heightened daily irritation and frustration this population experiences during daily life due to the difficulties dementia presents. It also plays a huge factor in whether the individual is willing to wear a new device or not. T his suggests that a device that needs to be placed on the individual every few days may be too obtrusive for certain people with dementia. Once the testing was complete, staff at Chelsea Place Memory Care that interacted with the fall detection system wer e asked to complete the system usability survey. The average score was 57 /100. The survey results show that the carestaff found the system easy to use and learn, not complex or cumbersome, and were somewhat confident while using it. Despite this, most woul d not like to use this system. The reason for this was because they had to continually check to make sure it was there and replace it in the morning if it fell off. It added additional steps to their already very busy morning schedule. These are important results because if the carestaff does not find the device useful it will not be used properly. If they system was able to stay on for multiple days, did not require the battery to be changed so often, and was smaller it may be more usable for the carestaf f. Future Fall Detection System Studies In the future this system needs to continually refine d, and the accuracy of the four threshold values needs to be improved This can happen by examining the current

PAGE 78

! *$ threshold test results, and by using a machine l earning technique. Once it is able to accurately detect which way the person fell it will help caretakers know where to look for injuries and help the users receive faster care. The algorithm also needs to be improved so that it can detect syncope falls. A syncope fall is defined as a fall where the user collapses against a wall but ends up sitting. The altitude data from these tests needs to be looked at further as well This needs to be completed in order to look at the varia bility of altitude changes d uring activities of daily living and to determine what would be a more acceptable threshold value. The data between the four activities of daily living versus a fall need to be compared further as well in order to better determine the differences between the two and improve the threshold values. This device also needs to be more compact so that the device and the battery can all fit easily within a Band Aid One way this can be achieved is by combining the two boards onto one surface mount printed circuit board and finding a smaller battery. This in addition to a slightly larger Band Aid, may also help the device stay on the residents throughout the night. Furthermore improvements need to be made to the battery life. Ideally, the device should last for up to a week so that caretakers are not constantly changing the battery. This can be accomplished by finding a smaller battery with higher amperage, and by finding ways to conserve power within the algorithm. This will also allow longer studies to be complet ed with the residents. The longer testing period will increase the likelihood of a resident falling while wearing the FDS and provide data to compare to the young adult study. In future iterations a smartphone application needs to be specifically designed for this system This will allow the system to be independent of a third party application send notifications faster, and work more efficiently within Anthem memory cares current

PAGE 79

! *% system A possible GPS locater or RFID tag would also be advantageous for t his device so that it would not be lost The device should also be waterproofed in the future to make it much more user friendly. This would also be beneficial so that the device could detect any falls that would occur in the shower or bathroom.

PAGE 80

! *& CHAPTER V CONCLUSION A real time accurate, wireless fall detection device that detect s falls that occur from a standing position by a resident with dementia in a memory care facility was developed. It combined a microcontroller, 10 Degree of Freedom IMU sensor, battery, WiFi, and a smartphone app lication Once this was created, h ealthy individuals between the ages of 20 5 6 complete 9 different falls and 4 activities of daily living. The calculated sensitivity, specificity, and accur acy for the device were 95%, 1 00 %, and 96% respectively These results validated the algorithm that was developed for the device Once this was complete, residents at Chelsea Place Memory Care were recruited to wear the mock device for 3 days and then the real device for an additional 3 days. This was to determine whether this device was properly designed for this population. It was found that this device is currently too large and falls off at night as a result. It was also found that the placement of the device on the lower shoulder caused little irritation or frustration to the residents throughout the day.

PAGE 81

! *' REFERENCES 1) Ortman, J., Velkoff, V., & Hogan, H. (2014). An Aging Nation: The Older Population in the United States Retrieved from https://www.census.gov/prod/2014pubs/p25 1140.pdf. 2) Wiener, Joshua, and Jane Tilly. "Population Ageing in the United States of America: Implications for Public Programmes." Int. J. Epidemiology2002. 776 81. Vol. 31 3) 2016 ALZHEIME R'S DISEASE FACTS AND FIGURES. (2016). Retrieved from http://www.alz.org/facts/overview.asp 4) Tabet, N., & Nelson, L. (2015). Slowing the progression of Alzheimer's disease; what works? (Vol. 23, pp. 19 3 209). Ageing Research Reviews. 5) Waite, L. (2015). Treatment for Alzheimer's disease: has anything changed? (Vol. 38, pp. 60 63). Australian Prescriber. 6) Barton, C., Ketelle, R., Merrilees, J., & Miller, B. (2016). Non pharmacological Management of Behavi oral Symptoms in Frontotemporal and Other Dementias (Vol. 16): Current Neurology and Neuroscience Reports. 7) Campdell, J., Robertson, C., Gardner, M., Norton, R., Tilyard, M., & Buchner, D. (1997). Randomised controlled trial of a general practice programme of home based exercise to prevent falls in elderly women 8) Rubenstein, L. (2006). Falls in older people: epidemiology, risk factors and strategies for prevention (Vol. 35, pp. 37 41). age and aging. 9) Eriksson, S., Gustafson, Y., & Lundin Olsson, L. (2008). Risk factors for falls in people with and without a diagnose of dementia living in residential care facilities: A prospective study (Vol. 46, pp. 293 306). Archives of Gerontology and Geriatrics. 10) Doorn, C. v., Gruber Baldini, A. L., Zimmerman, S., Hebel, J. R., Port, C., Baumgarten, M., Magaziner, J. (2003). Dementia as a Risk Factor for Falls and Fall Injuries Among Nursing Home Residents. Retrieved from http://www.medscape.com/viewarticle/46091 3 11) Tinetti, M. E., Liu, W. L., & Claus, E. B. (1993). Predictors and prognosis of inability to get up after falls among elderly persons (Vol. 269, pp. 65 70). J. Am. Med. Assoc. 12) Maki, E. B. (1997). Gait Changes in Older Adults: Predictors of Falls or In dicators of Fear? (Vol. 45, pp. 313 320): The American Geriatrics Society. 13) Gagnon, N., & Flint, A. J. (2003). Fear of Falling in the Elderly (Vol. 6, pp. 15 17). Geriatrics and Aging.

PAGE 82

! *( 14) Auvinet, E., Multon, F., Saint Arnaud, A., Rousseau, J., & Meunier, J. (2011). Fall Detection With Multiple Cameras: An Occlusion Resistant Method Based on 3 D 15) Belshaw, M., Taati, B., Snoek, J., & Mihailidis, A. (2011). Towards a single sensor passive solution for automated fall detection (pp. 1773 1776). 2011 Annual Inte rnational Conference of the IEEE Engineering in Medicine and Biology Society. 16) Sherwin, S., & Winsby, M. (2010). A relational perspective on autonomy for older adults residing in nursing homes (Vol. 14, pp. 182 190): Health Expectations. 17) Gasparrini, S., C ippitelli, E., Spinsante, S., & Gambi, E. (2014). A Depth Based Fall Detection System Using a Kinect¨ Sensor (Vol. 14, pp. 2756 2775). Sensors 18) Madansing, S., Thrasher, T., Layne, C., & Lee, B. C. (2015). Smartphone based fall detection system (pp. 370 374). Control, Automation and Systems (ICCAS), 2015 15th International Conference on. 19) He, Y., Li, Y., & Bao, S. D. (2012). Fall detection by built in tri accelerometer of smartphone (pp. 184 187). Proceedings of 2012 IEEE EMBS International Conference o n Biomedical and Health Informatics. 20) Fang, S. H., Liang, Y. C., & Chiu, K. M. (2012). Developing a mobile phone based fall detection system on Android platform (pp. 143 146). Computing, Communications and Applications Conference (ComComAp), 2012. 21) Li, Y ., Ho, K. C., Popescu, M., & Skubic, M. (2014). A Theoretical Study on The Placement of Microphone Arrays for Improving The Localization Accuracy of A Fall (pp. 4523 4526). 22) Li, Y., Popescu, M., Ho, K. C., & Nabelek, D. (2011). Improving Acoustic Fall Reco gnition by Adaptive Signal Windowing. 33rd Annual International Conference of the IEEE EMBS. 23) Li, Y., Zeng, Z., Popescu, M., & Ho, K. C. (2010). Acoustic Fall Detection Using a Circular Microphone Array (pp. 2242 2245). 32nd Annual International Conference of the IEEE EMBS. 24) Werner, F., Diermaier, J., Schmid, S., & Panek, P. (2011). Fall Detection with Distributed Floor mounted Accelerometers 25) Chaccour, K., Darazi, R., Hajjam el Hassani, A., & Andres, E. (2015). Smart Carpet using differential piezoresistiv e pressure sensors for elderly fall detection (pp. 225 229). 26) Aud, M., Abbott, C., Tyrer, H. W., Neelgund, R. V., Shriniwar, U. A. B., & Devarakonda, K. K. (2010). Smart Carpet: Developing a Sensor System to Detect Falls and Summon Assistance (pp. 8 12).

PAGE 83

! *) 27) Danes, S. (2012). Design for Dementia Care: A Retrospective Look at The Woodside Place Mode (Vol. 26, pp. 221 250): Journal of Housing for the Elderly. 28) Uegami, M., Iwamoto, T., & Matsumoto, M. (2012). A Study of Detection of Trip and Fall Using Doppler Se nsor on Embedded Computer (pp. 3263 3268). IEEE International Conference on Systems, Man, and Cybernetics. 29) Liu, L., Popescu, M., Skubic, M., Rantz, M., Yardibi, T., & Cuddihy, P. (2011). Automatic Fall Detection Based on Doppler Radar Motion Signature (pp 222 225). 5th International Conference on Pervasive Computing Technologies for Healthcare (PervasiveHealth) and Workshops. 30) Liu, L., Popescu, M., Skubic, M., & Rantz, M. (2014). An Automatic Fall Detection Framework Using Data Fusion of Doppler Radar and Motion Sensor Network (pp. 5940 5943). 31) Garripoli, C., Mercuri, M., Karsmakers, P., Soh, P. J., Crupi, G., Vandenbosch, G. A. E., Schreurs, D. (2015). Embedded DSP Based Telehealth Radar System for Remote In Door Fall Detection (Vol. 19, pp. 92 101) IEEE JOURNAL OF BIOMEDICAL AND HEALTH INFORMATICS. 32) Igual, R., Medrano, C., & Plaza, I. (2013). Challenges, issues and trends in fall detection systems (Vol. 12, pp. 1 24): BioMedical Engineering OnLine. 33) Bourke, A. K. Ven, P. v. d. Ga mble, M., O'Co nno r, R., Murphy, K., B ogan, E., Nelson, J. (2010). Applications of Waist Segment Kinematic Measurement using Accelerometry for an Autonomous Fall detection System during Continuous Activities (pp. 198 203). Signals and Systems Conference (ISSC 2010). 34) Chen, D., Feng, W., Zhang, Y., Li, X., & Wang, T. (2011). A Wearable Wireless Fall Detection System with Accelerators (pp. 2259 2263). International Conference on Robotics and Biomimetics. 35) Barton, C., Ketelle, R., Merrilees, J., & Miller, B. (2016). Non p harmacological Management of Behavioral Symptoms in Frontotemporal and Other Dementias (Vol. 16): Current Neurology and Neuroscience Reports. 36) Bharucha, J. S., Anand, V., Forlizzi, J., Dew, M. A., Reynolds III, F. C., Stevens, S., et al. (2009). Intelligen t Assistive Technology Applications to Dementia Care: Current Capabilities, Limitations, and Future Challenges (Vol. 17, pp. 88 104): The American Journal of Geriatric Psychiatry. 37) Igual, R., Medrano, C., & Plaza, I. (2013). Challenges, issues and trends i n fall detection systems (Vol. 12, pp. 1 24): BioMedical Engineering OnLine.

PAGE 84

! ** 38) Dumitrache, M., & Pasca, S. (2013). Fall Detection Algorithm Based on Triaxial Accelerometer Data (pp. 1 4). E Health and Bioengineering Conference (EHB). 39) Lan, C. C., Hsueh, Y. H., & Hu, R. Y. (2012). Real Time Fall Detecting System Using a Tri axial Accelerometer for Home Care (pp. 1077 1080). International Conference on Biomedical Engineering and Biotechnology. 40) Tolkiehn, M., Atallah, L., Lo, B., & Yang, G. Z. (2011). Direction Sensitive Fall Detection Using a Triaxial Accelerometer and a Barometric Pressure Sensor (pp. 369 372). 33rd Annual International Conference of the IEEE EMBS. 41) Bianchi, F., Redmond, S. J., Narayanan, M. R., Cerutti, S., & Lovell, N. H. (2010a). Barometric Pressure and Triaxial Accelerometry Based Falls Event Detection (Vol. 18, pp. 619 627). IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING. 42) Wang, C., Narayanan, M. R., Lord, S. R., Redmond, S. J., & Lovell, N. H. (2014). A Low power Fall Detection Algorithm Based on Triaxial Acceleration and Barometric Pressure (pp. 570 573). Conference Proceeding IEEE Engineering Medical Biology Society. 43) Pierleoni, P., Belli, A., Palma, L., Pellegrini, M., Pernini, L., & Valenti, S. (2015). A High Reliab ility Wearable Device for Elderly Fall Detection (Vol. 15, pp. 4544 4553). SENSORS 44) Onishi, J., Suzuki, Y., Umegaki, H., Endo, H., Kawamura, T., Imaizumi, M., et al. (2006). Behavioral, psychological and physical symptoms in group homes for older adults w ith dementia (pp. 75 86): International Psychogeriatric Association. '(D Rubenstein, L. (2006). Falls in older people: epidemiology, risk factors and strategies for prevention (Vol. 35, pp. 37 41). age and aging. 46) Karantonis, D. M., Narayanan, M. R., Mathie, M., Lovell, N. H., & Celler, B. G. (2006). Implementation of a real time human movement classifier using a triaxial accelerometer for ambulatory monitoring (Vol. 10, pp. 156 167). Trans. Inf. Technol. Biomed.

PAGE 85

! *+ APPENDIX A. Arduino Fall Detection C ode #include "Adafruit_MQTT.h" #include "Adafruit_MQTT_Client.h" #include #include #include #include #include #include #include #include #include #include #define AIO_SERVER "io.adafruit.com" #define AIO_SERVERPORT 8 883 // use 1883 8883 for SSL #define AIO_USERNAME "chrimack" #define AIO_KEY "d9f86 6da3bd74bf39ec7d83155e0d342" /* Assign a unique ID to the sensors */ Adafruit_10DOF dof = Adafruit_10DOF(); Adafruit_LSM303_Accel_Unified accel = Adafruit_LSM303_Accel_Unified(30301); Adafruit_LSM303_Mag_Unified mag = Adafruit_LSM303_ Mag_Unified(30302); Adafruit_BMP085_Unified bmp = Adafruit_BMP085_Unified(18001); /* Update this with the correct SLP for accurate altitude measurements */ float seaLevelPressure = SENSORS_PRESSURE_SEALEVELHPA; //float insTemp, desTemp, outTemp; / /char fall[100] = \ { \ "Roll \ ":orientation.roll, \ "Pitch \ ":orientation.pitch, \ "Altitude \ ":altitude1 \ \ "RMS \ ":rms1 \ \ "Fall \ ":f \ \ "Time \ ":time1 \ }"; char fall2[1000] = \ { \ "Roll2 \ ":R, \ "Pitch2 \ ":P, \ "Altitude2 \ ":A \ \ "RMS2 \ ":M \ \ "Fall \ ":F \ \ "Time \ :G \ }"; // Update these with values suitable for your network. //const char* ssid = "HoosierHouse"; const char* ssid = "Guest"; //const char* ssid = "Mackenzie Christensen"; //const char* password = "gopackgo"; const char* password = "willowBrook1"; //con st char* ssid = "More Beer, More Fun"; //const char* password = "DG860A336DC2";

PAGE 86

! *G const char* mqtt_server = "m11.cloudmqtt.com"; const char* id="abcd"; const char* usr="wfkyabjj"; const char* pass="6P1MAjsE0lm8"; WiFiClient client; Adafruit_MQTT_Client mqt t(&client, AIO_SERVER, AIO_SERVERPORT, AIO_USERNAME, AIO_KEY); Adafruit_MQTT_Publish fall3 = Adafruit_MQTT_Publish(&mqtt, AIO_USERNAME "/f/fall3"); WiFiClient wifiClient; PubSubClient mqttclient(wifiClient); //******************************************* ************************ *******/ void initSensors() { if(!accel.begin()) { /* There was a problem detecting the LSM303 ... check your connections */ Serial.println(F("Ooops, no LSM303 detected ... Check your wiring!")); fall3.publish("Accel orometer not connected"); while(1); } if(!mag.begin()) { /* There was a problem detecting the LSM303 ... check your connections */ Serial.println("Ooops, no LSM303 detected ... Check your wiring!"); fall3.publish("Magnetometer not con nected"); while(1); } if(!bmp.begin()) { /* There was a problem detecting the BMP180 ... check your connections */ Serial.println("Ooops, no BMP180 detected ... Check your wiring!"); fall3.publish("BMP180 not connected"); while(1) ; } } //*************************** Setup ************************************/

PAGE 87

! +H long lastMsg = 0; char msg[50]; int value = 0; float altitude1; float altitude2; float altitude3; float altitude4; float rms; float rms1; int myrms[1000]={ }; float myalt[1 000]={ }; int mypitch[1000]={ }; int myroll[1000]={ }; char* myfall[1000]={ }; int mytime[1000]={ }; int sample = 0; int J=0; int R=0; int M=0; int A=0; char* F=0; int P=0; int G=0; int roll=0; int pitch=0; void MQTT_connect(); void setup() { int A tI=2.5; //g int TtA=1; //s int AltA=.72; //g int AutA=1.28; //g int Otp=50; //degrees int Ttp=1; //s int rms2=0; int rms3=0; int rms4=0; int sample=0; pinMode(BUILTIN_LED, OUTPUT); // Initialize the BUILT IN_LED pin as an output Serial.begin(115200); setup_wifi(); mqttclient.setServer(mqtt_server, 2 6954); MQTT_connect(); }

PAGE 88

! +$ //*************************** Wifi ************************************/ void setup_wifi() { delay(10); Serial.println(); Serial.println(ssid); WiFi.begin(ssid, password); while (WiFi.status() != WL_CONNECTED) { delay(50); Serial.print("."); } Serial.println(""); Serial.println(WiFi.localIP()); /* Initialise the sensors */ initSensors(); MQTT_con nect(); } //*************************** MQTT ************************************/ void MQTT_connect() { int8_t ret; if (mqtt.connected()) { return; } uint8_t retries = 3; while ((ret = mqtt.connect()) != 0) { Serial.println(mqtt. connectErrorString(ret)); mqtt.disconnect(); delay(10); // wait 5 seconds retries -; if (retries == 0) { while (1); } } }

PAGE 89

! +% //*************************** Initial Loop ************************************/ void loop() { WiFi.forceSleepBegin(); delay(100); sensors_event_t accel_event; sensors_event_t mag_event; sensors_event_t bmp_event; sensors_vec_t orientation; /* Calculate pitch and roll from the raw accelerometer data */ accel.getEvent(& accel_event); float rms=abs((sqrt((sq(accel_event.acceleration.x))+(sq(accel_event.acceleration.y))+(sq( accel_event.acceleration.z)))) 10); accel.getEvent(&accel_event); if (dof.accelGetOrientation(&accel_event, &orientation)) { /* 'orientatio n' should have valid .roll and .pitch fields */ orientation.roll=orientation.roll; orientation.pitch=orientation.pitch; rms=rms; } //* Calculate the altitude using the barometric pressure sensor */ bmp.getEvent(&bmp_event); if (bm p_event.pressure) { /* Get ambient temperature in C */ float temperature; bmp.getTemperature(&temperature); /* Convert atmospheric pressure, SLP and temp to altitude */ float seaLevelPressure=SENSORS_PRESSURE_SEALEVELHPA; alti tude1=(bmp.pressureToAltitude(seaLevelPressure, bmp_event.pressure, temperature)); altitude1=a ltitude1; unsigned long time1=millis(); int my2darray[100][4]={ };

PAGE 90

! +& char* f="Normal"; myrms[sample]=rms; myalt[sample]=altitude1; myroll[sample]=orientation.roll; mypitch[sample]=orientation.pi tch; myfall[sample]=f; mytime[sample]=time1; } sample++; if (sample>=400){ sample=0; } delay(250); //*************************** Fall detection loop ************************************/ if (rms > 4) { for (int i=0; i <= 9; i++){ sensors_event_t accel_event; sensors_event_t mag_event; sensors_event_t bmp_event; sensors_vec_t orientation; /* Calculate pitch and roll from the raw accelerometer data */ accel.getEvent(&accel_event); float rms1=abs((sqrt((sq(accel_event.acceleration.x))+(sq(accel_event.acceleration.y))+(sq (accel_event.acceleration.z)))) 10); accel.getEvent(&accel_event); if (dof.accelGetOrientation(&accel_event, &orientation)) { /* 'orientation' should have valid .roll and .pitch fields */

PAGE 91

! +' orientation.roll= orientation.roll; int roll=orientation.roll; orientation.pitch=orientation.pitch; int pitch=orientation.pitch; rms1=rms1; } /* Cal culate the heading using the magnetometer */ /* Calculate the altitude using the barometric pressure sensor */ bmp.getEvent(&bmp_event); if (bmp_event.pressure) { /* Get ambient temperature in C */ float temperature; bmp.getTemperature(&temperature); /* Convert atmospheric pressure, SLP and temp to altitude */ altitude2=(bmp.pressureToAltitude(seaLevelPres sure, bmp_event.pressure, temperature)); altitude2=altitude2; unsigned long time1=millis(); int my2darray[100][4]={ }; char* q="RMS Triggered"; myrms[sample]=rms1; myalt[sample]=altitude2; myroll[sample]=orientation.rol l; mypitch[sample]=orientation.pitch; myfall[sample]=q; mytime[sample]=time1; sample++; }

PAGE 92

! +( if (sample>=400){ sample=0; } delay (10); if (1<=rms1<=2) //needs to check next rms event { float J = altitude2 altitude1; if (abs(J)>0.4) { delay(10); if (orientation.roll>= 25 && orientation.roll<=15 && orientation.pitch>= 25 && orientation.pitch<=15) { for (int o=0; o <= 100; o++){ sensors_event _t accel_event; sensors_event_t mag_event; sensors_event_t bmp_event; sensors_vec_t orientation; /* Calculate pitch and roll from the raw accelerometer data */ accel.getEvent(&accel_event); float rms1=abs((sqrt((sq(accel_event.acceleration.x))+(sq(accel_eve nt.acceleration.y))+(sq (accel_event.acceleration.z)))) 10); accel.getEvent(&accel_event); if (dof.accelGetOrientation(&accel_event, &orientation)) { /* 'orientation' should have valid .roll and .pitch fields */ orientation.roll=orientation.roll; roll=orie ntation.roll; orientation.pitch=orientation.pitch; pitch=orientation.pitch; rms1=rms1; // Serial.print(F("; ")); } /* Calculate the heading using the magnetometer */

PAGE 93

! +) //* Calculate the altitude using the barometric pressure sensor */ bmp.getEvent(&bmp_event); i f (bmp_event.pressure) { /* Get ambient temperature in C */ float temperature; bmp.getTemperature(&tempera ture); /* Convert atmospheric pressure, SLP and temp to altitude */ altitude3=(bmp.pressureToAltitude(seaLe velPressure, bmp_event.pressure, temperature)); altitude3=altitude3; unsigned long time1=millis(); int my2darray[100][4]={ }; char* n="Forward"; myrms[sample]=rms1; myalt[sample]=altitude3; myroll[sample]=orientation.roll; mypitch[sample]=orientation.pitch; myfal l[sample]=n; mytime[sample]=time1; sample++; } //Serial.print(sample++); if (sample>=400){ sample=0; } } WiFi.forceSleepWake(); delay(10);

PAGE 94

! +* Serial.println(); Serial.println(ssid); WiFi.begin(ssid, password); while (WiFi.status() != WL_CONNECTED) { delay(50); } Serial.println(""); Serial.println(WiFi.localIP()); MQTT_connect(); while (!mqttclient.connected()) { // Att empt to connect if (mqttclient.connect(id,usr,pass)) { // Once connected, publish an announcement... mqttclient.publish("topic", "he llo world"); // ... and resubscribe mqttclient.subscribe("inTopic"); } else { Serial.print (mqttclient.state()); // Wait 5 seconds before retrying delay(5000); }} initSensors(); fall3.publish("1"); fall3.publish("5"); for (int l=0; l < 400; l++){ StaticJsonBuffer<200> jsonBuffer; JsonObject& data =jsonBuffer.createObject(); float M=(myrms[l]); data["RMS"]=double_with_n_digits(M,4); float A=(myalt[l]); data["Altitude"]=double_with_n_digits(A,6); //Serial.print(A); R=(myroll[l]); data["Roll"]=R; P=(mypitch[l]);

PAGE 95

! ++ data["Pitch"]=P; F=(myfall[l]); data["Fall"]=F; G=(mytime[l]); data["Time"]=G; data.printTo(fall2, sizeof(fall2)); mqttclient.publish("fall2",fall2,2); //Serial.print(0); } fall3.publish("0"); } else if (orientation.roll >= 70 && orientation.roll<=50 && orientation.pitch>= 25 && orientation.pitch<=110) { //mqttclient.publish("fall3","Fall to the Right Detected",2); //fall3.publish("2"); for (int p=0; p <= 100; p++){ sensors_event_t accel_event; sensors_event_t mag_event; sens ors_event_t bmp_event; sensors_vec_t orientation; /* Calculate pitch and roll from the raw accelerometer data */ accel.getEvent(&accel_event); float rms1=abs((sqrt((sq(accel_event.acceleration.x))+(sq(accel_event.acceleration.y))+(sq (accel_event.acceleration.z)))) 10); accel.getEvent(&accel_e vent); if (dof.accelGetOrientation(&accel_event, &orientation)) { /* 'orientation' should have valid .roll and .pitch fields */ orientation.roll=orientation.roll; roll=orientation.roll; orientation.pitch=orientation.pi tch; pitch=orientation.pitch; rms1=rms1; // Serial.print(F("; ")); }

PAGE 96

! +G /* Calculate the heading using the magnetometer */ //* Calculate the altitude using the barometric pressure sensor */ bmp.getEvent(&bmp_event); if (bmp_event.pressure) { /* Get ambient temperature in C */ float temperatur e; bmp.getTemperature(&temperature); /* Convert atmospheric pressure, SLP and temp to altitude */ altitude4=(bmp.pressureToAltitude(seaLevelPressure, unsigned long time1=millis(); int my2darray[100][4]={ }; char* w="Right"; myrms[sample]=rms1; myalt[sample] =altitude4; myroll[sample]=orientation.roll; mypitch[sample]=orientation.pitch; myfall[sample]=w; mytime[sample]=time1; sample++; } if (sample>=400){ sample=0; } } WiFi.forceSleepWake(); delay(10); Serial.println(); //Serial.print("Con necting to "); Serial.println(ssid);

PAGE 97

! GH WiFi.begin(ssid, password); whi le (WiFi.status() != WL_CONNECTED) { delay(50); //Serial.print("."); } Serial.println(""); //Serial.println("WiFi connected 2"); //Serial.println("IP address: "); Serial.println(WiFi.localIP()); MQTT_connect(); while (!mqttclient.connected()) { // Attempt to connect if (mqttclient.connect(id,usr,pass)) { // Once connected, publish an announcement... mqttclient.publish("topic", "hello wor ld"); // ... and resubscribe mqttclient.subscribe("inTopic"); } else { //Serial.print("fai led, rc="); Serial.print(mqttclient.state()); //Serial.println(" try again in 5 seconds"); // Wait 5 seconds before retrying delay(5000); }} initSensors(); fall3.publish("2"); fall3.publish("5"); for (int y=0; y < 400; y++){ StaticJsonBuffer<200> jsonBuffer; JsonObject& data =jsonBuffer.createObject(); float M=(myrms[y]); data["RMS"]=double_with_n_digits(M,4); float A=(myalt[y]); data["Altitude"]=double_with_n_digits(A,6); Serial.print(A); R=(myroll[y]);

PAGE 98

! G$ data["Roll"]=R; P=(mypitch[y]); data["Pitch"]=P; F=(myfall[y]); data["Fall"]=F; G=(mytime[y]); data["Time"]=G; data.printTo(fall2, sizeof(fall2)); mqttclient.publish("fall2",fall2,2); } fall3. publish("0"); } else if (orientation.roll>= 20 && orientation.roll<=20 && orientation.pitch>= 190 && orientation.pitch<= 150) { for (int v=0; v <= 100; v++){ sensors_event_t accel_event; sensors_event_t mag_event; sensors_event_t bmp_event; sensors_vec_t orientation; /* Calculate pitch and roll from the raw accelerometer data */ accel.getEvent(&accel_ev ent); float rms1=abs((sqrt((sq(accel_event.acceleration.x))+(sq(accel_event.acceleration.y))+(sq (accel_event.acceleration.z)))) 10); accel.getEvent(&accel_event); if (dof.accelGetOrientation(&accel_event, &orientation)) { /* 'orientation' should have valid .roll and .pitch fields */ orientation.roll=orientation.roll; roll=orientation.roll; orientation.pitch=orientation.pitch; pitch=orientation.pitch; rms1=rms1; // Serial.print(F("; ")); }

PAGE 99

! G% /* Calculate the heading using the magnetometer */ //* Calculate the altitude using the barometric pressure sensor */ bmp.getEvent(&bmp_event); if (bmp_event.pressure) { /* Get ambient temperature in C */ float temperature; bmp.getTemperature(&temperature); /* Convert atmospheric pressure, SLP and temp to altitude */ float altitude4=(bmp.pressureToAltitude(seaLevelPressure, bmp_event.pressure, temperature)); altitude4=altitude4; unsigned long time1=millis(); int my2darray[100][4]={ }; char* a="Backwards"; myrms[sample]=rms1; myalt[sample]=altitude4; myroll[sample]=orientation.roll; mypitch[sample]=orientation.pitch; myfall[sample]=a; mytime[sample]=time1; sample++; } if (sample>=400){ sample =0; } } WiFi.forceSleepWake();

PAGE 100

! G& delay(10); Serial.println(); //Serial.print("Connecting to "); Serial.println(ssid); WiFi.begin(ssid, password); while (WiFi.status() != WL_CONNECTED) { delay(50); //Serial.print("."); } Serial.println(""); Serial.println(WiFi.localIP()); MQTT_connect(); while (!mqttclient.connected()) { // Attempt to connect if (mqttclient.connect(id,usr,pass)) { // Once connected, publish an announcement... mqttclient.publish("topic", "hello world"); // ... and resubscribe mqttclient.subscribe("inTopic"); } else { //Serial.print("failed, rc="); Serial.print(mqttclient.state()); //Serial.println(" try again in 5 seconds"); // Wait 5 seconds before retrying delay(5000); }} initSensors(); fall3.publish("3"); fall3.publish("5"); for (int h=0; h < 400; h++){ StaticJsonBuffer<200> jsonBuffer; JsonObject& data =jsonBuffer.cre ateObject(); float M=(myrms[h]); data["RMS"]=double_with_n_digits(M,4); float A=(myalt[h]); data["Altitude"]=double_with_n_digits(A,6);

PAGE 101

! G' Serial.print(A); R=(myroll[h]); data["Roll"]=R; P=(mypitch[h]); data["Pitch"]=P; F=(myfall[h]); data["Fall"]=F; G=(mytime[h]); data["Time"]=G; data.printTo(fall2, sizeof(fall2)); mqttclient.publish("fall2",fall2,2); //Serial.print(2); } fall3.publi sh("0"); } else if (orientation.roll>= 70 && orientation.roll<=25 && orientation.pitch>= 130 && orientation.pitch<= 60) { for (int t=0; t <= 100; t++){ sensors_event_t accel_event; sensors_event_t mag_event; sensors_event_t bmp_event; sensors_vec_t orientation; /* Calculate pitch and roll from the raw accelerometer data */ accel.getEvent(&accel_event); float rms1=abs((sqrt((sq(accel_event.acceleration.x))+(sq(accel_event.acceleration.y))+(sq (accel_event.acceleration.z)))) 10); accel.getEvent(&accel_event); if (dof.accelGetOrientation(&accel_event, &orientation)) { /* 'orientation' should have valid .roll and .pitch fields */ orientation.roll=orientation.roll; roll=orientation.roll; orientation.pitch=orientation.pitch; pitch=orientation.pitch; rms1=rms1;

PAGE 102

! G( } /* Calculate the heading us ing the magnetometer */ //* Calculate the altitude using the barometric pressure sensor */ bmp.getEvent(&bmp_event ); if (bmp_event.pressure) { /* Get ambient temperature in C */ float temperature; bmp.getTemperature(&temperature); /* Convert atmospheric pressure, SLP and temp to altitude */ float altitude5=(bmp.pressureToAltitude(seaLevelPressure, bmp_event.pressure, temperature)); altitude5=altitude5; unsigned long time1=millis(); int m y2darray[100][4]={ }; char* d="Left"; myrms[sample]=rms1; myalt[sample]=al titude5; myroll[sample]=orientation.roll; mypitch[sample]=orientation.pitch; myfall[sample]=d; mytime[sample]=time1; sample++; } if (sample>=400){ sample=0; } }

PAGE 103

! G) WiFi.forceSleepWake(); delay(10); Serial.println(); Serial.printl n(ssid); WiFi.begin(ssid, password); while (WiFi.status() != WL_CONNECTED) { delay(50); } Serial.println(""); Serial.println(WiFi.localIP()); MQTT_ connect(); while (!mqttclient.connected()) { // Attempt to connect if (mqttclient.connect(id,usr,pass)) { // Once connected, publish an announcement... mqttclient.publis h("topic", "hello world"); // ... and resubscribe mqttclient.subscribe("inTopic"); } else { //Serial.print("failed, rc="); Serial.print(mqttclient.state()); //Serial.println(" try again in 5 seconds"); // Wait 5 seconds before retrying delay(5000); }} initSensors(); fal l3.publish("4"); fall3.publish("5"); for (int c=0; c < 400; c++){ StaticJsonBuffer<200> jsonBuffer; JsonObject& data =jsonBuffer.createObject(); float M=(myr ms[c]); data["RMS"]=double_with_n_digits(M,4); float A=(myalt[c]);

PAGE 104

! G* data["Altitude"]=double_with_n_digits( A,6); Serial.print(A); R=(myroll[c]); data["Roll"]=R; P=(mypitch[c]); data["Pitch"]=P; F=(myfall[c]); data["Fall"]=F; G=(mytime[c]); data["Time"]=G; data.printTo(fall2, sizeof(fall2)); mqttclient.publish("fall2",f all2,2); } fall3.publish("0");} else { mqttclient.publish("fall3","no match",2); } } } delay(100); } } delay(200 ); // Wait for a second (previous 150) }

PAGE 105

! G+ B. Python Fall Data Collection Code import xlsxwriter import paho.mqtt.client as mqtt os, urlparse import json import time from datetime import datetime import csv import decimal now = str(datetime.now()) csvfile=open( 'test.csv' 'w' ) writer = csv.DictWriter(csvfile, fieldnames = [ "Time/Date" "Roll" "Pitch" "Altitude" "RMS" "Fall" "T ime" ]) writer.writeheader() csvfile.close() def on_connect(mqtt_client, userdata, rc): print ( "on_connect:: Connected with result code + str ( rc ) ) print ( "rc: + str(rc)) def on_message(mqtt_client, userdata, msg): print ( "on_message:: t his means I got a message from brokerfor this topic" ) print (msg.topic + + str(msg.qos) + + str(msg.payload)) global RMS global Roll global Pitch global Altitude global Time #global Battery #print(msg.payload.decode( 'utf 8')) json_data=json.loads(msg.payload) roll_data=json_data[ 'Roll' ] pitch_data=json_data[ 'Pitch' ] altitude_data=json_data[ 'Altitude' ] RMS_data=json_data[ 'RMS' ] fall_data=json_data[ 'Fall' ] time_data=json_data[ 'Time' ] #bat tery_data=json_data['Battery'] csvfile=open( 'test.csv' 'a' ) spamwriter = csv.writer(csvfile) spamwriter.writerow([now, roll_data, pitch_data, altitude_data, RMS_data, fall_data, time_data])

PAGE 106

! GG def on_publish(mqtt_client, userdata, m id): print ( "mid: + str(mid)) def on_subscribe(mqtt_client, userdata, mid, granted_qos): #print("This means broker has acknowledged my subscribe request") print ( "Subscribed: + str(mid) + + str(granted_qos)) def on_log(mqtt_client, user data, level, string): print (string) mqttc = mqtt.Client() # Assign event callbacks mqttc.on_message = on_message mqttc.on_connect = on_connect mqttc.on_publish = on_publish mqttc.on_subscribe = on_subscribe # Uncomment to enable debug messages mqt tc.on_log = on_log mqttc.username_pw_set( "wfkyabjj" "6P1MAjsE0lm8" ) mqttc.connect( 'm11.cloudmqtt.com' 2 6954 60 ) mqttc.loop_start() mqttc.subscribe ( "fall2" 2 ) run = True while run: mqttc.publish ( "/toclientloud" "from python code" ) #cli ent.subscribe ("fall" ,0 ) mqttc.subscribe ( "fall2 2 ) #client.publish ( "/frommothership", "on") time.sleep( 2 ) #client.publish ( "/frommothership", "off") #time.sleep(2)

PAGE 107

! $HH C. Arduino Code to Test Battery Life Constantly Connected to Wifi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c $HHd!]!e f g f e`0;; f eh03".-464"0-=30;;B!! f eM"4/5 f eh03".-464"0-=<"4/5B! f eF;4"489. f eh6;4"489.$! f B!! f e`TL f eh312$ f B! f eO6;; f eh? f B! f e>"1. f eh4"1.$ f ie^ /563!?6;;%c$HHHd!]!e f g f e`0;;% f eh`B!! f eM"4/5% f ehMB! f eF;4"489.% f ehF! f B!! f e`TL% f ehT f B! f eO6;; f ehO f B! f e>"1. f ehW f B! f eV644 .37 f ehbie^ ZZ!U<964.!45.2.!@"45!#6;8.2!28"46:;.!?03!7083!-.4@03j= ZZ/0-24!/563[!22"9!]!eb002".3b082.e^ /0-24!/563[!22"9!]!eT6/j.-k".!Q53"24.-2.-e^ /0-24!/563[!<622@039!]!eA0<6/jA0e^ ZZ/0-24!/563[!22"9!]!eT03.!V..3B!T03.!O8-e^ ZZ/0-24!/563[!<622@039!]!eXW +)HF&&)XQ%e^ /0-24!/563[!1E44R2.3#.3!]!e1$$=/;0891E44=/01e^ /0-24!/563[!"9]e6:/9e^ /0-24!/563[!823]e@?j76:lle^ /0-24!/563[!<622]e)M$TFl2KH;1+e^ N"O"Q;".-4!@"?"Q;".-4^ M8:L8:Q;".-4!1E44/;".-4C@"?"Q;".-4D^ ZZ[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[ [[[[[[[[[[[[[[[[[[[[[[[[Z #0"9!"-"4L.-2032CD g !! "?Cm6//.;=:.A"-CDD !! g

PAGE 108

! $H$ !!!! Z[!>5.3.!@62!6!<30:;.1!9.4./4"-A!45.!SLT&H&!===!/5./j!7083!/0--./4"0-2![Z !!!! L.3"6;=<3"-4;-COCe,00<2B!-0!SLT&H&!9.4./4.9!===!Q5./j!7083!@"3"-AmeDD^ !!!! @5";.C$D^ !! i !! "?Cm16A=:.A "-CDD !! g !!!! Z[!>5.3.!@62!6!<30:;.1!9.4./4"-A!45.!SLT&H&!===!/5./j!7083!/0--./4"0-2![Z !!!! L.3"6;=<3"-4;-Ce,00<2B!-0!SLT&H&!9.4./4.9!===!Q5./j!7083!@"3"-AmeD^ !!!! @5";.C$D^ !! i !! "?Cm:1<=:.A"-CDD !! g !!!! Z[!>5.3.!@62!6!<30:;.1!9.4./4"-A!45.!VTM$+H!===!/5. /j!7083!/0--./4"0-2![Z !!!! L.3"6;=<3"-4;-Ce,00<2B!-0!VTM$+H!9.4./4.9!===!Q5./j!7083!@"3"-AmeD^ !!!! @5";.C$D^ !! i i ZZ[[[[[[[[[[[[[[[[[[[[[[[[[[[!L.48
PAGE 109

! $H% !!!!! "-4!312&]H^ !!!!! "-4!312']H^ !! <"-T09.CVU\S>\_RSKXB!,U>MU>D^!!!!!ZZ!\-"4"6;"k.!45.!VU\S>\_RSKX!<"-!62!6-! 084<84 !! L.3"6;=:.A"-C$$(%HHD^ !! 2.48KXD!g !!!! 9.;67C(HHD^ !!!! L.3"6;=<3"-4Ce=eD^ !! i !! L.3"6;=<3"-4;-CeeD^ !! L.3"6;=<3"-4;-CeN"O"!/0--./4.9eD^ !! L.3"6;=<3"-4;-Ce\M!6993.22h!eD^ !! L.3"6;=<3 "-4;-CN"O"=;0/6;\MCDD^ !! Z[!\-"4"6;"2.!45.!2.-2032![Z !! "-"4L.-2032CD^ i ZZ[[[[[[[[[[[[[[[[[[[[[[[[[[[!Tn>>![[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[Z #0"9!/6;;:6/jC/563[!40<"/B!:74.[!<67;069B!8-2"A-.9!"-4!;.-A45D!g !! ZZL.3"6;=<3"-4CeT.226A.!633"#.9!ceD^ !! ZZL .3"6;=<3"-4C40<"/D^ !! ZZL.3"6;=<3"-4Ced!eD^ !! ?03!C"-4!"!]!H^!"!J!;.-A45^!"ooD!g !!!! ZZL.3"6;=<3"-4CC/563D<67;069c"dD^ !! i !! ZZL.3"6;=<3"-4;-CD^ !! ZZ!L@"4/5!0-!45.!SKX!"?!6-!$!@62!3./."#.9!62!?"324!/5636/4.3 !! ZZ"?!CC/563D<67;069cHd!]]!p$pD!g

PAGE 110

! $H& !!!! ZZ9"A"46 ;N3"4.CVU\S>\_RSKXB!S,ND^!!!ZZ!>83-!45.!SKX!0-!C_04.!4564!S,N!"2!45.! #0;46A.!;.#.; !!!! ZZ!:84!6/486;;7!45.!SKX!"2!0-^!45"2!"2!:./682. !!!! ZZ!"4!"2!6/"#.!;0@!0-!45.!KLM q H$D !! ZZi!.;2.!g !!!! ZZ9"A"46;N3"4.CVU\S>\_RSKXB!b\WbD^!!ZZ!>83-!45.!SKX!0??!:7!16j"-A!4 5.!#0;46A.! b\Wb !! i ! ZZ8"-4&%R4!r]H^ #0"9!3./0--./4CD!g !! ZZ!S00>!/0--./4"0-===eD^ !!!! ZZ!F44.1<4!40!/0--./4 !!!! "?!C1E44/;".-4=/0--./4C"9B823B<622DD!g !!!!!! L .3"6;=<3"-4;-Ce/0--./4.9eD^ !!!!!! ZZ!,-/.!/0--./4.9B!<8:;"25!6-!6--08-/.1.-4=== !!!!!! 1E44/;".-4=<8:;"25Ce40<"/eB!e5.;;0!@03;9eD^ !!!!!! ZZ!===!6-9!3.28:2/3":. !!!!!! 1E44/;".-4=28:2/3":.Ce"->0<"/eD^ !!!! i!.;2.!g !!!!!! L.3"6;=<3"-4Ce?6";.9B!3/]eD^ !!!!!! L.3" 6;=<3"-4C1E44/;".-4=2464.CDD^ !!!!!! L.3"6;=<3"-4;-Ce!437!6A6"-!"-!(!2./0-92eD^ !!!!!! ZZ!N6"4!(!2./0-92!:.?03.!3.437"-A !!!!!! 9.;67C(HHHD^ !!!! i !! i i ZZ[[[[[[[[[[[[[[[[[[[[[[[[[[[!\-"4"6;!S00
PAGE 111

! $H' !! Z[!Q6;/8;64.!<"4/5!6-9!30;;!?301!45.!36@!6//.;.301.4.3!9646![Z !! 6//.;=A.4K#.-4Cs6//.;R.#.-4D^ !!! "-4! 312]6:2CC2E34CC2EC6//.;R.#.-4=6//.;.364"0-=rDDoC2EC6//.;R.#.-4=6//.;.364"0-=7DDoC2 EC6//.;R.#.-4=6//.;.364"0-=kDDDD q $HD^ !! 6//.;=A.4K#.-4Cs6//.;R.#.-4D^ !! "?!C90?=6//.;W.4,3".-464"0-Cs6//.;R.#.-4B!s03".-464"0-DD !! g !!! Z[!p03".-464"0-p!2508;9!56#.!#6;"9!=30;;!6-9!=<"4/5!?".;92![Z !!!! 03".-464"0-=30;;]03".-464"0-=30;;^ !!!! 03".-464"0-=<"4/5]03".-464"0-=<"4/5^ !!!! 312]312^ ZZ!!!!L.3"6;=<3"-4COCe^!eDD^ !! i !! !! Z[!Q6;/8;64.!45.!5.69"-A!82"-A!45.!16A-.401.4.3![Z !! ZZ[! Q6;/8;64.!45.!6;4"489.!82"-A!45.!:6301.43"/!<3.2283.!2.-203![Z !! :1<=A.4K#.-4Cs:1.1<.36483.Cs4.1<.36483.D^ !!!! Z[!Q0-#.34!64102<5.3"/!<3.2283 .B!LSM!6-9!4.10F;4"489.C2.6S.#.;M3.2283.B !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! :1
PAGE 112

! $H( !!! 174"1.c$d]4"1.$^ !!! 17:644.37c$d];.#.;^ !! !!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! ZZN"O"=:.A"-C22"9B!<622@039D^ !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! L464"/Y20-V8??.3J%HHP!l20-V8??.3^ !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! Y20-,:l./4s!9646!]l20-V8??.3=/3.64.,:l./4CD^ !! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! T]C17312c$dD^ !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 9646ce`TLed]T^ !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! ?;064!F]C176;4c$dD^ !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 9646ceF; 4"489.ed]908:;.R@"45R-R9"A"42CFB)D^ !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! `]C1730;;c$dD^ !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 9646ce`0;;ed]`^ !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! M]C17<"4/5c$dD^ !!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!! 9646ceM"4/5ed]M^ !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! O]C17?6;;c$dD^ !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 9646ceO6;;ed]O^ !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! W]C174"1.c$dD^ !!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 9646ce>"1.ed]W^ !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! b]C17:644.37c$dD^ !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 9646ceV644.37ed]b^ !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! ZZL.3"6;=<3"-4 C17:644.37c;dD^ !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! ZZL.3"6;=<3"-4C261<;.%D^ !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 9646=<3"-4>0C?6;;%B!2"k.0?C?6;;%DD^ !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 1E44/;".-4=<8:;"25Ce?6;;%eB? 6;;%B%D^ !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! ZZL.3"6;=<3"-4CHD^ !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!! i!!!9.;67C$HHHD^!!i

PAGE 113

! $H) D Arduino Code to Test Battery Life Only Connecting to WiF i When Fall Triggered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c$HHd!]!e f g f e`0;; f eh03".-464"0-=30;;B!! f eM"4/5 f eh03".-464"0-=<"4/5B! f eF;4"489. f eh6;4"489.$! f B!! f e`TL f eh312$ f B! f eO6;; f eh? f B! f e>"1. f eh4"1.$ f ie^ /563!?6;;%c$HHHd!]!e f g f e`0;;% f eh`B!! f eM"4/5% f ehMB! f eF;4"489.% f ehF! f B!! f e`TL% f ehT f B! f eO6;; f ehO f B! f e>"1 f ehW f B! f eV644.37 f ehbie^ ZZ!U<964.!45.2.!@"45!#6;8.2!28"46:;.!?03!7083!-.4@03j= /0-24!/563[!22"9!]!eb002".3b082.e^ ZZ/0-24!/563[!22"9!]!eT6/j.-k".!Q53"24.-2.-e^ /0-24!/563[!<622@039!]!eA0<6/jA0e^ ZZ/0-24!/563[!22"9!]!eT03.!V..3B!T03.!O8-e^ ZZ/0-24!/563[! <622@039!]!eXW+)HF&&)XQ%e^ /0-24!/563[!1E44R2.3#.3!]!e1$$=/;0891E44=/01e^ /0-24!/563[!"9]e6:/9e^ /0-24!/563[!823]e@?j76:lle^ /0-24!/563[!<622]e)M$TFl2KH;1+e^ N"O"Q;".-4!@"?"Q;".-4^ M8:L8:Q;".-4!1E44/;".-4C@"?"Q;".-4D^ ZZ[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[ [[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[Z #0"9!"-"4L.-2032CD g !! "?Cm6//.;=:.A"-CDD !! g

PAGE 114

! $H* !!!! Z[!>5.3.!@62!6!<30:;.1!9.4./4"-A!45.!SLT&H&!===!/5./j!7083!/0--./4"0-2![Z !!!! L.3"6;=<3"-4;-COCe,00<2B!-0!SLT&H&!9.4./4.9!===!Q5./j!7083!@"3"-AmeDD^ !!!! @5";.C$D^ !! i !! "?Cm16A=:.A"-CDD !! g !!!! Z[!>5.3.!@62!6!<30:;.1!9.4./4"-A!45.!SLT&H&!===!/5./j!7083!/0--./4"0-2![Z !!!! L.3"6;=<3"-4;-Ce,00<2B!-0!SLT&H&!9.4./4.9!===!Q5./j!7083!@"3"-AmeD^ !!!! @5";.C$D^ !! i !! "?Cm:1<=:.A"-CDD !! g !!!! Z[!>5.3.!@62!6!<30:;.1!9.4./4"-A!45.! VTM$+H!===!/5./j!7083!/0--./4"0-2![Z !!!! L.3"6;=<3"-4;-Ce,00<2B!-0!VTM$+H!9.4./4.9!===!Q5./j!7083!@"3"-AmeD^ !!!! @5";.C$D^ !! i i ZZ[[[[[[[[[[[[[[[[[[[[[[[[[[[!L.48
PAGE 115

! $H+ !!!!! "-4!312&]H^ !!!!! "-4!312']H^ !! <"-T09.CVU\S>\_RSKXB!,U>MU>D^!!!!!ZZ!\-"4"6;"k.!45.!VU\S>\_RSKX!<"-!62!6-! 084<84 !! L.3"6;=:.A"-C$$(%HHD^ !! 2.48KXD!g !!!! 9.;67C(HHD^ !!!! L.3"6;=<3"-4Ce=eD^ !! i !! L.3"6;=<3"-4;-CeeD^ !! L.3"6;=<3"-4;-CeN"O"!/0--./4.9eD^ L.3"6;=<3"-4;-Ce\M!6993.22h!eD^ !! L.3"6;=<3"-4;-CN"O"=;0/6;\MCDD^ !! Z[!\-"4"6;"2.!45.!2.-2032![Z !! "-"4L.-2032CD^ i ZZ[[[[[[[[[[[[[[[[[[[[[[[[[[[!\-"4"6;!S00
PAGE 116

! $HG !! 2.-2032R.#.-4R4!6//.;R.#.-4^ !! 2.-2032R.#.-4R4!16AR.#.-4^ !! 2.-2032R.#.-4R4!:1.1<.36483.Cs4.1<.36483.D^ !!!! Z[!Q0-#.34!6410 2<5.3"/!<3.2283.B!LSM!6-9!4.10F;4"489.C2.6S.#.;M3.2283.B !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! :1
PAGE 117

! $$H !!! 176;4c$d]6;4"489.$^ !!! 1730;;c$d]03".-464"0-=30;;^ !!! 17<"4/5c$d];.#.;%^ !!! 174"1.c$d]4"1.$^ !!! 17:644.37c$d];.#.;^ N"O"=?03/.L;..KXD!g !!!! 9.;67C(HHD^ !!!! L.3"6;=<3"-4Ce=eD^ !! i !! L.3"6;=<3"-4;-CeeD^ L.3"6;=<3"-4;-CeN"O"!/0--./4.9!%eD^ !! L.3"6;=<3"-4;-Ce\M!6993.22h!eD^ !! L.3"6;=<3"-4;-CN"O"=;0/6;\MCDD^ !! @5";.!Cm1E44/;".-4=/0--./4.9CDD!g !!!! L.3"6;=<3"-4CeF44.1<4"-A!Tn>>!/0--./4"0-===eD^ !!!! ZZ!F44.1<4!40!/0--./4 !!!! "?!C1E44/;".-4=/0--./4C"9B823B<6 22DD!g !!!!!! L.3"6;=<3"-4;-Ce/0--./4.9eD^ !!!!!! ZZ!,-/.!/0--./4.9B!<8:;"25!6-!6--08-/.1.-4=== !!!!!! 1E44/;".-4=<8:;"25Ce40<"/eB!e5.;;0!@03;9eD^ !!!!!! ZZ!===!6-9!3.28:2/3":. !!!!!! 1E44/;".-4=28:2/3":.Ce"->0<"/eD^ !!!! i!.;2.!g !!!!!! L.3"6;=<3"-4Ce?6";.9B!3/] eD^ !!!!!! L.3"6;=<3"-4C1E44/;".-4=2464.CDD^ !!!!!! L.3"6;=<3"-4;-Ce!437!6A6"-!"-!(!2./0-92eD^ !!!!!! ZZ!N6"4!(!2./0-92!:.?03.!3.437"-A !!!!!! 9.;67C(HHHD^ !!!! ii !!!!!!

PAGE 118

! $$$ ZZN"O"=:.A"-C22"9B!<622@039D^ !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! L464"/Y20V8??.3J%HHP!l20-V8??.3^ !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! Y20-,:l./4s!9646!]l20-V8??.3=/3.64.,:l./4CD^ !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! T]C17312c$dD^ !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 9646ce`TLed]T^ !!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! ?;064!F]C176;4c$dD^ !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 9646ceF;4"489.ed]908:;.R@"45R-R9"A"42CFB)D^ !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! `]C1730;;c$dD^ !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!! 9646ce`0;;ed]`^ !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! M]C17<"4/5c$dD^ !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 9646ceM"4/5ed]M^ !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! O]C17?6;;c$dD^ !!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!! 9646ceO6;;ed]O^ !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! W]C174"1.c$dD^ !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 9646ce>"1.ed]W^ !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! b]C17:644.37c$dD^ !!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 9646ceV644.37ed]b^ !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! ZZL.3"6;=<3"-4C17:644.37c;dD^ !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! ZZL.3"6;=<3"-4C261<;.%D^ !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!! 9646=<3"-4>0C?6;;%B!2"k.0?C?6;;%DD^ !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 1E44/;".-4=<8:;"25Ce?6;;%eB?6;;%B%D^ !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! ZZL.3"6;=<3"-4CHD^ !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!! i!!!9.;67C(HHHD^!!i

PAGE 119

! $$% E Python Code Collecting Sensor Data import xlsxwriter import paho.mqtt.client as mqtt os, urlparse import json import time import csv import decimal now = time.strftime( '%d %m %Y %H:%M: %S' ) csvfile=open( 'test.csv' 'w' ) writer = csv.DictWriter(csvfile, fieldnames = [ "Time/Date" "Roll" "Pitch" "Altitude" "RMS" "Fall" "Time" ]) writer.writeheader() csvfile.close() def on_connect(mqtt_client, userdata, rc): print ( "on_connect:: Co nnected with result code + str ( rc ) ) print ( "rc: + str(rc)) def on_message(mqtt_client, userdata, msg): print ( "on_message:: this means I got a message from brokerfor this topic" ) print (msg.topic + + str(msg.qos) + + str(msg.payl oad)) global RMS global Roll global Pitch global Altitude global Time #global Battery #print(msg.payload.decode('utf 8')) json_data=json.loads(msg.payload) roll_data=json_data[ 'Roll' ] pitch_data=json_data[ 'Pitch' ] altitude_data=json_data[ 'Altitude' ] RMS_data=json_data[ 'RMS' ] fall_data=json_data[ 'Fall' ] time_data=json_data[ 'Time' ] #battery_data=json_data['Battery'] csvfile=open( 'test.csv' 'a' ) spamwriter = csv.writer(csvfile) spamwr iter.writerow([now, roll_data, pitch_data, altitude_data, RMS_data, fall_data, time_data])

PAGE 120

! $$& def on_publish(mqtt_client, userdata, mid): print ( "mid: + str(mid)) def on_subscribe(mqtt_client, userdata, mid, granted_qos): #print("This mean s broker has acknowledged my subscribe request") print ( "Subscribed: + str(mid) + + str(granted_qos)) def on_log(mqtt_client, userdata, level, string): print (string) mqttc = mqtt.Client() # Assign event callbacks mqttc.on_message = on_mess age mqttc.on_connect = on_connect mqttc.on_publish = on_publish mqttc.on_subscribe = on_subscribe # Connect mqttc.username_pw_set( "wfkyabjj" "6P1MAjsE0lm8" ) mqttc.connect( 'm11.cloudmqtt.com' 16954 60 ) # Start subscribe, with QoS level 0 mqttc.subscri be( "fall2" 2 ) # Continue the network loop, exit when an error occurs rc = 0 while rc == 0 : rc = mqttc.loop() print ( "rc: + str(rc))

PAGE 121

! $$' F. Python Code Collecting Battery Data import xlsxwriter import paho.mqtt.client as mqtt os, urlparse import jso n import time import csv import decimal now = time.strftime( '%d %m %Y %H:%M:%S' ) csvfile=open( 'test.csv' 'w' ) writer = csv.DictWriter(csvfile, fieldnames = [ "Time/Date" "Roll" "Pitch" "Altitude" "RMS" "Fall" "Time" "Battery" ]) writer.writeheader() csvfile.close() def on_connect(mqtt_client, userdata, rc): print ( "on_connect:: Connected with result code + str ( rc ) ) print ( "rc: + str(rc)) def on_message(mqtt_client, userdata, msg): print ( "on_message:: this means I got a message f rom brokerfor this topic" ) print (msg.topic + + str(msg.qos) + + str(msg.payload)) global RMS global Roll global Pitch global Altitude global Time global Battery #print(msg.payload.decode('utf 8')) json_data=json. loads(msg.payload) roll_data=json_data[ 'Roll' ] pitch_data=json_data[ 'Pitch' ] altitude_data=json_data[ 'Altitude' ] RMS_data=json_data[ 'RMS' ] fall_data=json_data[ 'Fall' ] time_data=json_data[ 'Time' ] battery_data=json_data[ 'Battery' ] csvfile=open( 'test.csv' 'a' ) spamwriter = csv.writer(csvfile) spamwriter.writerow([now, roll_data, pitch_data, altitude_data, RMS_data, fall_data, time_data, battery_data])

PAGE 122

! $$( def on_publish(mqtt_client, userdata, mid): print ( mid: + str(mid)) def on_subscribe(mqtt_client, userdata, mid, granted_qos): #print("This means broker has acknowledged my subscribe request") print ( "Subscribed: + str(mid) + + str(granted_qos)) def on_log(mqtt_client, userdata, level, str ing): print (string) mqttc = mqtt.Client() # Assign event callbacks mqttc.on_message = on_message mqttc.on_connect = on_connect mqttc.on_publish = on_publish mqttc.on_subscribe = on_subscribe # Uncomment to enable debug messages mqttc.on_log = on_lo g mqttc.username_pw_set( "wfkyabjj" "6P1MAjsE0lm8" ) mqttc.connect( 'm11.cloudmqtt.com' 16954 60 ) mqttc.loop_start() mqttc.subscribe ( "fall2" 2 ) run = True while run: mqttc.publish ( "/tomqttcloud" "from python code" ) #client.subscribe ("fa ll" ,0 ) mqttc.subscribe ( "fall2 2 ) #client.publish ( "/frommothership", "on") time.sleep( 2 ) #client.publish ( "/frommothership", "off") #time.sleep(2)

PAGE 123

! $$) G Fall Backward End on Left Shoulder Graphs for Young 375 380 385 390 395 400 Time (ms) -100 -80 -60 -40 -20 0 20 Degrees (¡) Subplot Roll 375 380 385 390 395 400 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 375 380 385 390 395 400 Time (ms) 1580 1582 1584 1586 1588 Estimated Altitude (m) Subplot Altitude 375 380 385 390 395 400 Time (ms) 0 5 10 15 20 Acceleration (g) Subplot RMS Figure 22 Example of resulting roll, pitch, altitude, and RMS values from subject 11 falling backwards and end on their left shoulder. 430 435 440 445 450 455 Time (ms) -80 -60 -40 -20 0 20 Degrees (¡) Subplot Roll 430 435 440 445 450 455 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 430 435 440 445 450 455 Time (ms) 1578 1579 1580 1581 1582 1583 1584 Estimated Altitude (m) Subplot Altitude 430 435 440 445 450 455 Time (ms) 0 5 10 15 20 25 Acceleration (g) Subplot RMS Figure 23 Example of resulting roll, pitch, altitude, and RMS values from subject 13 falling backwards and end on their left shoulder.

PAGE 124

! $$* 325 330 335 340 345 350 Time (ms) -100 -80 -60 -40 -20 0 20 Degrees (¡) Subplot Roll 325 330 335 340 345 350 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 325 330 335 340 345 350 Time (ms) 251 252 253 254 255 256 Estimated Altitude (m) Subplot Altitude 325 330 335 340 345 350 Time (ms) 0 1 2 3 4 5 6 Acceleration (g) Subplot RMS Figure 24 Example of resulting roll, pitch, altitude, and RMS values from subject 15 falling backwards and end on their left shoulder. 290 300 310 320 330 Time (ms) -100 -80 -60 -40 -20 0 Degrees (¡) Subplot Roll 290 300 310 320 330 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 290 300 310 320 330 Time (ms) 251 252 253 254 255 256 257 Estimated Altitude (m) Subplot Altitude 290 300 310 320 330 Time (ms) 0 2 4 6 8 Acceleration (g) Subplot RMS Figure 25 Example of resulting roll, pitch, altitude, and RMS values from subject 17 falling backwards and end on their left shoulder.

PAGE 125

! $$+ H Fall Forward End on Left Shoulder Graphs 160 170 180 190 Time (ms) -80 -60 -40 -20 0 20 40 Degrees (¡) Subplot Roll 160 170 180 190 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 160 170 180 190 Time (ms) 1579 1580 1581 1582 1583 1584 Estimated Altitude (m) Subplot Altitude 160 170 180 190 Time (ms) 0 2 4 6 8 10 Acceleration (g) Subplot RMS Figure 26 Example of resulting roll, pitch, altitude, and RMS values from subject 11 falling forwards and end on their left shoulder. 190 200 210 220 230 Time (ms) -100 -50 0 50 Degrees (¡) Subplot Roll 190 200 210 220 230 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 190 200 210 220 230 Time (ms) 1540 1560 1580 1600 1620 Estimated Altitude (m) Subplot Altitude 190 200 210 220 230 Time (ms) 0 2 4 6 8 10 12 Acceleration (g) Subplot RMS Figure 27 Example of resulting roll, pitc h, altitude, and RMS values from subject 13 falling forwards and end on their left shoulder.

PAGE 126

! $$G 300 305 310 315 320 Time (ms) -100 -50 0 50 100 Degrees (¡) Subplot Roll 300 305 310 315 320 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 300 305 310 315 320 Time (ms) 205 210 215 Estimated Altitude (m) Subplot Altitude 300 305 310 315 320 Time (ms) 0 2 4 6 8 10 Acceleration (g) Subplot RMS Figure 28 Example of resulting roll, pitch, altitude, and RMS values from subject 14 falling forwards and end on their left shoulder.

PAGE 127

! $%H I Lateral Left Fall Graphs 790 795 800 805 810 815 Time (ms) -100 -80 -60 -40 -20 0 20 Degrees (¡) Subplot Roll 790 795 800 805 810 815 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 790 795 800 805 810 815 Time (ms) 1588 1589 1590 1591 1592 1593 1594 Estimated Altitude (m) Subplot Altitude 790 795 800 805 810 815 Time (ms) 0 2 4 6 8 10 Acceleration (g) Subplot RMS Figure 29 Example of resulting roll, pitch, altitude, and RMS values from subject 9 lateral fall left. 840 850 860 870 Time (ms) -80 -60 -40 -20 0 20 Degrees (¡) Subplot Roll 840 850 860 870 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 840 850 860 870 Time (ms) 1586 1588 1590 1592 1594 1596 Estimated Altitude (m) Subplot Altitude 840 850 860 870 Time (ms) 0 5 10 15 20 Acceleration (g) Subplot RMS Figure 30 Example of resulting roll, pitch, altitude, and RMS values from subject 10 lateral fall left.

PAGE 128

! $%$ 100 110 120 130 140 Time (ms) -100 -80 -60 -40 -20 0 20 Degrees (¡) Subplot Roll 100 110 120 130 140 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 100 110 120 130 140 Time (ms) 1578 1579 1580 1581 1582 1583 1584 Estimated Altitude (m) Subplot Altitude 100 110 120 130 140 Time (ms) 0 2 4 6 8 Acceleration (g) Subplot RMS Figure 31 Example of resulting r oll, pitch, altitude, and RMS values from subject 11 lateral fall left. 590 595 600 605 610 Time (ms) -100 -50 0 50 Degrees (¡) Subplot Roll 590 595 600 605 610 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 590 595 600 605 610 Time (ms) 1580 1582 1584 1586 1588 1590 1592 Estimated Altitude (m) Subplot Altitude 590 595 600 605 610 Time (ms) 0 2 4 6 8 Acceleration (g) Subplot RMS Figure 32 Example of resulting roll, pitch, altitude, and RMS values from subject 12 lateral fall left.

PAGE 129

! $%% 5 10 15 20 25 30 Time (ms) -100 -50 0 50 100 Degrees (¡) Subplot Roll 5 10 15 20 25 30 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 5 10 15 20 25 30 Time (ms) 1579 1580 1581 1582 1583 1584 1585 Estimated Altitude (m) Subplot Altitude 5 10 15 20 25 30 Time (ms) 0 2 4 6 8 Acceleration (g) Subplot RMS Figure 33 Example of resulting roll, pitch, altitude, and RMS values fro m subject 13 lateral fall left. 110 120 130 140 Time (ms) -100 -50 0 50 Degrees (¡) Subplot Roll 110 120 130 140 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 110 120 130 140 Time (ms) 209 210 211 212 213 214 215 Estimated Altitude (m) Subplot Altitude 110 120 130 140 Time (ms) 0 2 4 6 8 10 12 Acceleration (g) Subplot RMS Figure 34 Example of resulting roll, pitch, altitude, and RMS values from subject 14 lateral fall left.

PAGE 130

! $%& 710 715 720 725 730 735 Time (ms) -40 -20 0 20 40 60 80 Degrees (¡) Subplot Roll 710 715 720 725 730 735 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 710 715 720 725 730 735 Time (ms) 248 249 250 251 252 253 Estimated Altitude (m) Subplot Altitude 710 715 720 725 730 735 Time (ms) 0 1 2 3 4 5 6 Acceleration (g) Subplot RMS Figure 35 Example of resulting roll, pitch, altitude, and RMS values from subject 15 lateral fall left. 255 260 265 270 275 280 Time (ms) -60 -40 -20 0 20 40 60 Degrees (¡) Subplot Roll 255 260 265 270 275 280 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 255 260 265 270 275 280 Time (ms) 252 254 256 258 260 262 264 Estimated Altitude (m) Subplot Altitude 255 260 265 270 275 280 Time (ms) 0 2 4 6 8 10 Acceleration (g) Subplot RMS Figur e 36 Example of resulting roll, pitch, altitude, and RMS values from subject 18 lateral fall left.

PAGE 131

! $%' J Fall Forward End on Right Shoulder Graphs 130 140 150 160 170 Time (ms) -100 -50 0 50 100 Degrees (¡) Subplot Roll 130 140 150 160 170 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 130 140 150 160 170 Time (ms) 1589 1590 1591 1592 1593 1594 1595 Estimated Altitude (m) Subplot Altitude 130 140 150 160 170 Time (ms) 0 2 4 6 8 10 Acceleration (g) Subplot RMS Figure 37 Example of resulting roll, pitch, altitude, and RMS values from subject 9 falling forward and en d on their right shoulder. 120 130 140 150 160 Time (ms) -80 -60 -40 -20 0 Degrees (¡) Subplot Roll 120 130 140 150 160 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 120 130 140 150 160 Time (ms) 1589 1590 1591 1592 1593 1594 1595 Estimated Altitude (m) Subplot Altitude 120 130 140 150 160 Time (ms) 0 5 10 15 20 Acceleration (g) Subplot RMS Figure 38 Example of resulting roll, pitch, altitude, and RMS values from subject 10 falling forward and end on their right shoulder.

PAGE 132

! $%( 145 150 155 160 165 Time (ms) -100 -50 0 50 100 Degrees (¡) Subplot Roll 145 150 155 160 165 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 145 150 155 160 165 Time (ms) 1580 1581 1582 1583 1584 1585 1586 Estimated Altitude (m) Subplot Altitude 145 150 155 160 165 Time (ms) 0 2 4 6 8 10 12 Acceleration (g) Subplot RMS Figure 39 Example of resulting roll, pitch, altitude, and RMS values from subject 12 fa lling forward and end on their right shoulder. 190 195 200 205 210 Time (ms) -100 -50 0 50 100 Degrees (¡) Subplot Roll 190 195 200 205 210 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 190 195 200 205 210 Time (ms) 249 249.5 250 250.5 251 251.5 Estimated Altitude (m) Subplot Altitude 190 195 200 205 210 Time (ms) 0 2 4 6 8 10 Acceleration (g) Subplot RMS Figure 40 Example of resulting roll, pitch, altitude, and RMS values from subject 16 falling forward and end on their right shoulder.

PAGE 133

! $%) 150 155 160 165 Time (ms) -100 -50 0 50 Degrees (¡) Subplot Roll 150 155 160 165 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 150 155 160 165 Time (ms) 242 244 246 248 250 252 254 Estimated Altitude (m) Subplot Altitude 150 155 160 165 Time (ms) 0 5 10 15 20 Acceleration (g) Subplot RMS Figure 41 Example of resulting roll, pitch, altitude, and RMS value s from subject 17 falling forward and end on their right shoulder. 60 70 80 90 100 110 Time (ms) -100 -50 0 50 Degrees (¡) Subplot Roll 60 70 80 90 100 110 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 60 70 80 90 100 110 Time (ms) 250 252 254 256 258 260 Estimated Altitude (m) Subplot Altitude 60 70 80 90 100 110 Time (ms) 0 5 10 15 20 Acceleration (g) Subplot RMS Figure 42 Example of resulting roll, pitch, altitude, and RMS values from subject 18 falling forward and end on their right shoulder.

PAGE 134

! $%* K Fall Backwards End on Right Shoulder Graphs 420 430 440 450 460 Time (ms) -80 -60 -40 -20 0 20 Degrees (¡) Subplot Roll 420 430 440 450 460 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 420 430 440 450 460 Time (ms) 1586 1588 1590 1592 1594 1596 Estimated Altitude (m) Subplot Altitude 420 430 440 450 460 Time (ms) 0 5 10 15 20 Acceleration (g) Subplot RMS F igure 43 Example of resulting roll, pitch, altitude, and RMS values from subject 10 falling backwards and end on their right shoulder. 350 360 370 380 390 400 Time (ms) -100 -50 0 50 Degrees (¡) Subplot Roll 350 360 370 380 390 400 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 350 360 370 380 390 400 Time (ms) 1581 1582 1583 1584 1585 1586 1587 Estimated Altitude (m) Subplot Altitude 350 360 370 380 390 400 Time (ms) 0 2 4 6 8 10 Acceleration (g) Subplot RMS Figure 44 Example of resulting roll, pitch, altitude, and RMS values from subject 12 falling backwards and end on th eir right shoulder.

PAGE 135

! $%+ 360 380 400 420 Time (ms) -100 -50 0 50 100 Degrees (¡) Subplot Roll 360 380 400 420 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 360 380 400 420 Time (ms) 205 206 207 208 209 210 211 Estimated Altitude (m) Subplot Altitude 360 380 400 420 Time (ms) 0 2 4 6 8 10 Acceleration (g) Subplot RMS Figure 45 Example of resulting roll, pitch, altitude, and RMS values from subject 14 falling backwards and end on their right shoulder. 400 410 420 430 Time (ms) -100 -80 -60 -40 -20 0 Degrees (¡) Subplot Roll 400 410 420 430 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 400 410 420 430 Time (ms) 250.5 251 251.5 252 252.5 253 Estimated Altitude (m) Subplot Altitude 400 410 420 430 Time (ms) 0 2 4 6 8 10 12 Acceleration (g) Subplot RMS Figure 46 Example of resulting roll, pitch, altitude, and RMS values from subject 16 falling backwards and end on their right shoulder.

PAGE 136

! $%G L. Lateral Fall Right Graphs 585 590 595 600 605 Time (ms) -100 -80 -60 -40 -20 0 20 Degrees (¡) Subplot Roll 585 590 595 600 605 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 585 590 595 600 605 Time (ms) 1589 1590 1591 1592 1593 1594 1595 Estimated Altitude (m) Subplot Altitude 585 590 595 600 605 Time (ms) 0 5 10 15 20 Acceleration (g) Subplot RMS Figure 47 Example of resulting roll, pitch, altitude, and RMS values from subject 9 fall lateral right. 640 645 650 655 660 665 Time (ms) -100 -50 0 50 Degrees (¡) Subplot Roll 640 645 650 655 660 665 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 640 645 650 655 660 665 Time (ms) 1588 1590 1592 1594 1596 Estimated Altitude (m) Subplot Altitude 640 645 650 655 660 665 Time (ms) 0 5 10 15 Acceleration (g) Subplot RMS Figure 48 Example of resulting roll, pitch, altitude, and RMS values fr om subject 10 fall lateral right.

PAGE 137

! $&H 440 450 460 470 480 Time (ms) -100 -50 0 50 Degrees (¡) Subplot Roll 440 450 460 470 480 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 440 450 460 470 480 Time (ms) 1580 1582 1584 1586 1588 1590 Estimated Altitude (m) Subplot Altitude 440 450 460 470 480 Time (ms) 0 5 10 15 20 Acceleration (g) Subplot RMS Figure 49 Example of resulting roll, pitch, altitude, and RMS values from subject 11 fall lateral right. 765 770 775 780 785 790 Time (ms) -100 -50 0 50 100 Degrees (¡) Subplot Roll 765 770 775 780 785 790 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 765 770 775 780 785 790 Time (ms) 1580 1582 1584 1586 1588 Estimated Altitude (m) Subplot Altitude 765 770 775 780 785 790 Time (ms) 0 2 4 6 8 10 Acceleration (g) Subplot RMS Figure 50 Example of resulting roll, pitch, altitude, and RMS values from subject 12 fall lateral right.

PAGE 138

! $&$ 210 220 230 240 250 Time (ms) -100 -50 0 50 Degrees (¡) Subplot Roll 210 220 230 240 250 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 210 220 230 240 250 Time (ms) 1578 1580 1582 1584 1586 Estimated Altitude (m) Subplot Altitude 210 220 230 240 250 Time (ms) 0 2 4 6 8 10 12 Acceleration (g) Subplot RMS F igure 51 Example of resulting roll, pitch, altitude, and RMS values from subject 13 fall lateral right. 575 580 585 590 595 Time (ms) -80 -60 -40 -20 0 20 40 Degrees (¡) Subplot Roll 575 580 585 590 595 Time (ms) 0 50 100 150 200 Degrees (¡) Subplot Pitch 575 580 585 590 595 Time (ms) 207 208 209 210 211 212 Estimated Altitude (m) Subplot Altitude 575 580 585 590 595 Time (ms) 0 2 4 6 8 10 12 Acceleration (g) Subplot RMS Figure 52 Example of resulting roll, pitch, altitude, and RMS values from subject 14 fall lateral right.

PAGE 139

! $&% 520 530 540 550 560 Time (ms) 0 20 40 60 80 100 Degrees (¡) Subplot Roll 520 530 540 550 560 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 520 530 540 550 560 Time (ms) 248 249 250 251 252 253 Estimated Altitude (m) Subplot Altitude 520 530 540 550 560 Time (ms) 0 2 4 6 8 Acceleration (g) Subplot RMS Figure 53 Example of resulting roll, p itch, altitude, and RMS values from subject 15 fall lateral right. 560 565 570 575 580 585 Time (ms) -100 -80 -60 -40 -20 0 20 Degrees (¡) Subplot Roll 560 565 570 575 580 585 Time (ms) -50 0 50 100 150 200 Degrees (¡) Subplot Pitch 560 565 570 575 580 585 Time (ms) 246 248 250 252 254 256 258 Estimated Altitude (m) Subplot Altitude 560 565 570 575 580 585 Time (ms) 0 5 10 15 20 Acceleration (g) Subplot RMS Figure 54 Example of resulting roll, pitch, altitude, and RMS values from subject 17 fall lateral right.

PAGE 140

! $&& 90 95 100 105 110 115 Time (ms) -100 -50 0 50 100 Degrees (¡) Subplot Roll 90 95 100 105 110 115 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 90 95 100 105 110 115 Time (ms) 248 250 252 254 256 258 260 Estimated Altitude (m) Subplot Altitude 90 95 100 105 110 115 Time (ms) 0 2 4 6 8 10 Acceleration (g) Subplot RMS Figure 55 Example of resulting roll, pitch, altitude, and RMS values from su bject 18 fall lateral right.

PAGE 141

! $&' M Lateral Fall Left End on Stomach Graphs 860 870 880 890 900 910 Time (ms) -100 -50 0 50 Degrees (¡) Subplot Roll 860 870 880 890 900 910 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 860 870 880 890 900 910 Time (ms) 1588 1590 1592 1594 1596 1598 1600 Estimated Altitude (m) Subplot Altitude 860 870 880 890 900 910 Time (ms) 0 5 10 15 20 Acceleration (g) Subplot RMS Figure 56 Example of resulting roll, pitch, altitude, and RMS values from subject 9 lateral fall left end on stomach. 955 960 965 970 975 Time (ms) -80 -60 -40 -20 0 20 Degrees (¡) Subplot Roll 955 960 965 970 975 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 955 960 965 970 975 Time (ms) 1586 1587 1588 1589 1590 1591 1592 Estimated Altitude (m) Subplot Altitude 955 960 965 970 975 Time (ms) 0 5 10 15 Acceleration (g) Subplot RMS Figure 57 Example of resulting roll, pitch, altitude, and RMS values from subject 10 lateral fall left end on stomach.

PAGE 142

! $&( 310 315 320 325 330 Time (ms) -100 -50 0 50 Degrees (¡) Subplot Roll 310 315 320 325 330 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 310 315 320 325 330 Time (ms) 1575 1580 1585 1590 1595 Estimated Altitude (m) Subplot Altitude 310 315 320 325 330 Time (ms) 0 2 4 6 8 10 12 Acceleration (g) Subplot RMS Figure 58 Example of resulting roll, pitch, altitude, and RMS values from subject 11 lateral fall left end on stomach. 660 670 680 690 700 Time (ms) -80 -60 -40 -20 0 20 40 Degrees (¡) Subplot Roll 660 670 680 690 700 Time (ms) -50 0 50 100 150 200 Degrees (¡) Subplot Pitch 660 670 680 690 700 Time (ms) 1581 1582 1583 1584 1585 1586 1587 Estimated Altitude (m) Subplot Altitude 660 670 680 690 700 Time (ms) 0 2 4 6 8 10 12 Acceleration (g) Subplot RMS Figure 59 Example of resulting roll, pitch, altitude, and RMS values from subject 12 lateral fall left end on stomach.

PAGE 143

! $&) 115 120 125 130 135 140 Time (ms) -80 -60 -40 -20 0 20 40 Degrees (¡) Subplot Roll 115 120 125 130 135 140 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 115 120 125 130 135 140 Time (ms) 1578 1580 1582 1584 1586 1588 Estimated Altitude (m) Subplot Altitude 115 120 125 130 135 140 Time (ms) 0 1 2 3 4 5 Acceleration (g) Subplot RMS Figure 60 Example of resulting roll, pitch, altitude, and RMS values from subject 13 lateral fall left end on stomach. 180 190 200 210 Time (ms) -80 -70 -60 -50 -40 -30 -20 Degrees (¡) Subplot Roll 180 190 200 210 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 180 190 200 210 Time (ms) 208 210 212 214 216 Estimated Altitude (m) Subplot Altitude 180 190 200 210 Time (ms) 0 5 10 15 20 Acceleration (g) Subplot RMS Figure 61 Example of resulting roll, pitch, altitude, and RMS values fr om subject 14 lateral fall left end on stomach.

PAGE 144

! $&* 780 800 820 840 Time (ms) 0 20 40 60 80 100 Degrees (¡) Subplot Roll 780 800 820 840 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 780 800 820 840 Time (ms) 248 249 250 251 252 253 254 Estimated Altitude (m) Subplot Altitude 780 800 820 840 Time (ms) 0 2 4 6 8 10 Acceleration (g) Subplot RMS Figure 62 Example of resulting roll, pitch, altitude, and RMS values from subject 15 lateral fall left end on stomach. 730 735 740 745 750 755 Time (ms) -80 -60 -40 -20 0 20 Degrees (¡) Subplot Roll 730 735 740 745 750 755 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 730 735 740 745 750 755 Time (ms) 250 250.5 251 251.5 252 252.5 Estimated Altitude (m) Subplot Altitude 730 735 740 745 750 755 Time (ms) 0 2 4 6 8 Acceleration (g) Subplot RMS Figure 63 Example of resulting roll, pitch, altitude, and RMS values from subject 16 lateral fall left end on stomach.

PAGE 145

! $&+ 470 480 490 500 510 Time (ms) -100 -80 -60 -40 -20 0 20 Degrees (¡) Subplot Roll 470 480 490 500 510 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 470 480 490 500 510 Time (ms) 252 253 254 255 256 257 Estimated Altitude (m) Subplot Altitude 470 480 490 500 510 Time (ms) 0 5 10 15 20 Acceleration (g) Subplot RMS Figure 64 Example of resulting roll, pitch, altitude, and RMS values from subject 17 lateral fall left end on stomach.

PAGE 146

! $&G N Lateral Right Fall End on Stomach Graphs 670 675 680 685 690 Time (ms) -100 -50 0 50 Degrees (¡) Subplot Roll 670 675 680 685 690 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 670 675 680 685 690 Time (ms) 1589 1590 1591 1592 1593 Estimated Altitude (m) Subplot Altitude 670 675 680 685 690 Time (ms) 0 5 10 15 Acceleration (g) Subplot RMS Figure 65 Example of resulting roll, pitch altitude, and RMS values from subject 9 lateral fall right end on stomach. 720 730 740 750 760 Time (ms) -80 -60 -40 -20 0 Degrees (¡) Subplot Roll 720 730 740 750 760 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 720 730 740 750 760 Time (ms) 1588 1589 1590 1591 1592 1593 Estimated Altitude (m) Subplot Altitude 720 730 740 750 760 Time (ms) 0 5 10 15 20 Acceleration (g) Subplot RMS Figure 66 Example of resulting roll, pitch, altitude, and RMS values from subject 10 lateral fall left end on stomach.

PAGE 147

! $'H 530 540 550 560 570 Time (ms) -100 -80 -60 -40 -20 0 20 Degrees (¡) Subplot Roll 530 540 550 560 570 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 530 540 550 560 570 Time (ms) 1580 1585 1590 1595 Estimated Altitude (m) Subplot Altitude 530 540 550 560 570 Time (ms) 0 5 10 15 20 Acceleration (g) Subplot RMS Figure 67 Example of resulting roll, pitch, altitude and RMS values from subject 11 lateral fall left end on stomach. 840 850 860 870 880 Time (ms) -100 -50 0 50 Degrees (¡) Subplot Roll 840 850 860 870 880 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 840 850 860 870 880 Time (ms) 1580 1582 1584 1586 1588 Estimated Altitude (m) Subplot Altitude 840 850 860 870 880 Time (ms) 0 5 10 15 20 25 Acceleration (g) Subplot RMS Figure 68 Example of resulting roll, pitch, altitude, and RMS values from subject 12 lateral fall left end on stomach.

PAGE 148

! $'$ 335 340 345 350 Time (ms) -100 -50 0 50 100 Degrees (¡) Subplot Roll 335 340 345 350 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 335 340 345 350 Time (ms) 1579 1580 1581 1582 1583 1584 1585 Estimated Altitude (m) Subplot Altitude 335 340 345 350 Time (ms) 0 5 10 15 20 25 Acceleration (g) Subplot RMS Figure 69 Example of resulting roll, pitch, altitude, and RMS values from subject 13 lateral fall left end on stomach. 32 34 36 38 Time (ms) -100 -80 -60 -40 -20 0 20 Degrees (¡) Subplot Roll 32 34 36 38 Time (ms) -20 -10 0 10 20 Degrees (¡) Subplot Pitch 32 34 36 38 Time (ms) 208 209 210 211 212 213 214 Estimated Altitude (m) Subplot Altitude 32 34 36 38 Time (ms) 0 5 10 15 20 Acceleration (g) Subplot RMS Figure 70 Example of resulting roll, pitch, altitude, and RMS values from subject 14 lateral fall left end on stomach.

PAGE 149

! $'% 620 630 640 650 660 Time (ms) -20 0 20 40 60 80 100 Degrees (¡) Subplot Roll 620 630 640 650 660 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 620 630 640 650 660 Time (ms) 248 249 250 251 252 253 Estimated Altitude (m) Subplot Altitude 620 630 640 650 660 Time (ms) 0 1 2 3 4 Acceleration (g) Subplot RMS Figure 71 Example of resulting roll, pitch, altitude, and RMS values fro m subject 15 lateral fall left end on stomach. 610 620 630 640 650 Time (ms) -100 -50 0 50 Degrees (¡) Subplot Roll 610 620 630 640 650 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 610 620 630 640 650 Time (ms) 250 250.5 251 251.5 252 252.5 Estimated Altitude (m) Subplot Altitude 610 620 630 640 650 Time (ms) 0 2 4 6 8 10 Acceleration (g) Subplot RMS Figure 72 Example of resulting roll, pitch, altitude, and RMS values from subject 16 lateral fall left end on stomach.

PAGE 150

! $'& 675 680 685 690 695 Time (ms) -80 -60 -40 -20 0 20 40 Degrees (¡) Subplot Roll 675 680 685 690 695 Time (ms) -50 0 50 100 150 200 Degrees (¡) Subplot Pitch 675 680 685 690 695 Time (ms) 252 253 254 255 256 257 Estimated Altitude (m) Subplot Altitude 675 680 685 690 695 Time (ms) 0 5 10 15 20 Acceleration (g) Subplot RMS Figure 73 Example of resulting roll, pitch, altitude, and RMS values from subject 17 lateral fall left end on stomach. 150 160 170 180 190 Time (ms) -100 -50 0 50 100 Degrees (¡) Subplot Roll 150 160 170 180 190 Time (ms) -50 0 50 100 150 200 Degrees (¡) Subplot Pitch 150 160 170 180 190 Time (ms) 254 255 256 257 258 259 260 Estimated Altitude (m) Subplot Altitude 150 160 170 180 190 Time (ms) 0 5 10 15 Acceleration (g) Subplot RMS Figure 74 Example of resulting roll, pitch, altitude, and RMS values from subject 18 lateral fall left end on stomach.

PAGE 151

! $'' O Fall Forward Graphs 0 20 40 60 80 Time (ms) -100 -50 0 50 Degrees (¡) Subplot Roll 0 20 40 60 80 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 0 20 40 60 80 Time (ms) 1590 1592 1594 1596 1598 Estimated Altitude (m) Subplot Altitude 0 20 40 60 80 Time (ms) 0 2 4 6 8 10 12 Acceleration (g) Subplot RMS Figure 75 Example of resulting roll, pitch, altitude, and RMS values from subject 9 falling forward. 10 20 30 40 50 60 Time (ms) -100 -80 -60 -40 -20 0 Degrees (¡) Subplot Roll 10 20 30 40 50 60 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 10 20 30 40 50 60 Time (ms) 1588 1590 1592 1594 1596 Estimated Altitude (m) Subplot Altitude 10 20 30 40 50 60 Time (ms) 0 2 4 6 8 Acceleration (g) Subplot RMS Figure 76 Example of resulting roll, pitch, altitude, and RMS values from subject 10 falling forward.

PAGE 152

! $'( 40 50 60 70 Time (ms) -100 -50 0 50 Degrees (¡) Subplot Roll 40 50 60 70 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 40 50 60 70 Time (ms) 1578 1580 1582 1584 1586 Estimated Altitude (m) Subplot Altitude 40 50 60 70 Time (ms) 0 2 4 6 8 Acceleration (g) Subplot RMS Figure 77 Example of resulting roll, pitch, altitude, and RMS values from subject 11 falling forward. 10 20 30 40 50 60 Time (ms) -100 -50 0 50 100 Degrees (¡) Subplot Roll 10 20 30 40 50 60 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 10 20 30 40 50 60 Time (ms) 1580 1582 1584 1586 1588 Estimated Altitude (m) Subplot Altitude 10 20 30 40 50 60 Time (ms) 0 2 4 6 8 10 Acceleration (g) Subplot RMS Figure 7 8 Example of resulting roll, pitch, altitude, and RMS values from subject 12 falling forward.

PAGE 153

! $') 70 75 80 85 90 Time (ms) -80 -60 -40 -20 0 Degrees (¡) Subplot Roll 70 75 80 85 90 Time (ms) -200 -150 -100 -50 0 50 Degrees (¡) Subplot Pitch 70 75 80 85 90 Time (ms) 1578 1580 1582 1584 1586 Estimated Altitude (m) Subplot Altitude 70 75 80 85 90 Time (ms) 0 5 10 15 20 Acceleration (g) Subplot RMS Figure 79 Example of resulting roll, pitch, altitude, and RMS values from subject 13 falling forward. 130 140 150 160 Time (ms) -100 -80 -60 -40 -20 0 20 Degrees (¡) Subplot Roll 130 140 150 160 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 130 140 150 160 Time (ms) 248 250 252 254 256 Estimated Altitude (m) Subplot Altitude 130 140 150 160 Time (ms) 0 2 4 6 8 Acceleration (g) Subplot RMS Figure 80 Example of resulting roll, pitch, altitud e, and RMS values from subject 15 falling forward.

PAGE 154

! $'* 50 60 70 80 90 100 Time (ms) -100 -80 -60 -40 -20 0 Degrees (¡) Subplot Roll 50 60 70 80 90 100 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 50 60 70 80 90 100 Time (ms) 248.5 249 249.5 250 250.5 251 251.5 Estimated Altitude (m) Subplot Altitude 50 60 70 80 90 100 Time (ms) 0 5 10 15 Acceleration (g) Subplot RMS Figure 81 Example of resulting roll, pitch, altitude, and RMS values from subject 16 falling forward. 55 60 65 70 75 Time (ms) -100 -50 0 50 Degrees (¡) Subplot Roll 55 60 65 70 75 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 55 60 65 70 75 Time (ms) 249 250 251 252 253 254 Estimated Altitude (m) Subplot Altitude 55 60 65 70 75 Time (ms) 0 5 10 15 Acceleration (g) Subplot RMS Figure 82 Example of resulting roll, pitch, altitude, and RMS values from subject 17 falling fo rward.

PAGE 155

! $'+ 5 10 15 20 25 Time (ms) -100 -80 -60 -40 -20 0 Degrees (¡) Subplot Roll 5 10 15 20 25 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 5 10 15 20 25 Time (ms) 252 253 254 255 256 257 Estimated Altitude (m) Subplot Altitude 5 10 15 20 25 Time (ms) 0 5 10 15 Acceleration (g) Subplot RMS Figure 83 Example of resulting roll, pitch, altitude, and RMS values from subject 18 falling forward.

PAGE 156

! $'G P Fall Backwar d s, Land in Sitting Position, End on Back Graphs 465 470 475 480 485 490 Time (ms) -100 -80 -60 -40 -20 0 Degrees (¡) Subplot Roll 465 470 475 480 485 490 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 465 470 475 480 485 490 Time (ms) 1586 1588 1590 1592 1594 1596 Estimated Altitude (m) Subplot Altitude 465 470 475 480 485 490 Time (ms) 0 1 2 3 4 5 6 Acceleration (g) Subplot RMS Figure 84 Example of resulting roll, pitch, altitude, and RMS values fr om subject 9 ?6;;! : 6/j@63 92B!;6-9!"-!2"44"-A!<02"4"0-B!.-9!0-!: 6/j 540 550 560 570 Time (ms) -100 -80 -60 -40 -20 0 20 Degrees (¡) Subplot Roll 540 550 560 570 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 540 550 560 570 Time (ms) 1560 1580 1600 1620 1640 1660 Estimated Altitude (m) Subplot Altitude 540 550 560 570 Time (ms) 0 2 4 6 8 10 12 Acceleration (g) Subplot RMS Figure 85 Example of resulting roll, pitch, altitude, and RMS values from subject 10 ?6;;! : 6/j@63 92B!;6-9!"-!2"44"-A!<02"4"0-B!.-9!0-!: 6/j

PAGE 157

! $(H 10 15 20 25 Time (ms) -100 -80 -60 -40 -20 0 Degrees (¡) Subplot Roll 10 15 20 25 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 10 15 20 25 Time (ms) 1575 1580 1585 1590 Estimated Altitude (m) Subplot Altitude 10 15 20 25 Time (ms) 0 5 10 15 Acceleration (g) Subplot RMS Figure 86 Example of resulting roll, pi tch, altitude, and RMS values from subject 11 ?6;;! : 6/j@63 92B!;6-9!"-!2"44"-A!<02"4"0-B!.-9!0-!: 6/j 490 495 500 505 510 515 Time (ms) -100 -80 -60 -40 -20 0 20 Degrees (¡) Subplot Roll 490 495 500 505 510 515 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 490 495 500 505 510 515 Time (ms) 1540 1550 1560 1570 1580 1590 1600 Estimated Altitude (m) Subplot Altitude 490 495 500 505 510 515 Time (ms) 0 2 4 6 8 10 Acceleration (g) Subplot RMS Figure 87 Example of resulting roll, pitch, altitude, and RMS values from subject 12 ?6;;! : 6/j@63 92B!;6-9!"-!2"44"-A!<02"4"0-B!.-9!0-!: 6/j

PAGE 158

! $($ 40 50 60 70 80 Time (ms) -100 -80 -60 -40 -20 0 20 Degrees (¡) Subplot Roll 40 50 60 70 80 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 40 50 60 70 80 Time (ms) 1560 1565 1570 1575 1580 1585 1590 Estimated Altitude (m) Subplot Altitude 40 50 60 70 80 Time (ms) 0 2 4 6 8 10 Acceleration (g) Subplot RMS Figure 88 Example of resulting roll, pitch, altitude, and RMS values from subject 13 ?6;;! : 6/j@63 92B!;6-9!"-!2"44"-A!<02"4"0-B!.-9!0-!: 6/j 500 505 510 515 520 525 Time (ms) -100 -80 -60 -40 -20 0 20 Degrees (¡) Subplot Roll 500 505 510 515 520 525 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 500 505 510 515 520 525 Time (ms) 205 206 207 208 209 210 211 Estimated Altitude (m) Subplot Altitude 500 505 510 515 520 525 Time (ms) 0 5 10 15 20 Acceleration (g) Subplot RMS Figure 89 Example of resulting roll, pitch, altitude, and RMS values from subject 14 ?6;;! : 6/j@63 92B!;6-9!"-!2"44"-A <02"4"0-B!.-9!0-!: 6/j

PAGE 159

! $(% 410 420 430 440 Time (ms) -20 0 20 40 60 80 100 Degrees (¡) Subplot Roll 410 420 430 440 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 410 420 430 440 Time (ms) 210 220 230 240 250 260 Estimated Altitude (m) Subplot Altitude 410 420 430 440 Time (ms) 0 1 2 3 4 5 Acceleration (g) Subplot RMS Figure 90 Example of resulting roll, pitch, altitude, and RMS values from subject 15 ?6;;! : 6/j@63 92B!;6-9!"-!2"44"-A!<02"4"0-B!.-9!0-!: 6/j 500 505 510 515 520 Time (ms) -100 -80 -60 -40 -20 0 20 Degrees (¡) Subplot Roll 500 505 510 515 520 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 500 505 510 515 520 Time (ms) 240 250 260 270 280 Estimated Altitude (m) Subplot Altitude 500 505 510 515 520 Time (ms) 0 1 2 3 4 Acceleration (g) Subplot RMS Figure 91 Example of resulting roll, pitch, altitude, and RMS values from subject 16 ?6;;! : 6/j@63 92B!;6-9!"-!2"44"-A!<02"4"0-B!.-9!0-!: 6/j

PAGE 160

! $(& 385 390 395 400 405 410 Time (ms) -100 -50 0 50 Degrees (¡) Subplot Roll 385 390 395 400 405 410 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 385 390 395 400 405 410 Time (ms) 220 230 240 250 260 270 Estimated Altitude (m) Subplot Altitude 385 390 395 400 405 410 Time (ms) 0 5 10 15 20 Acceleration (g) Subplot RMS Figure 92 Example of resulting roll, pitch, altitude, and RMS values from subject 17 ?6;;! : 6/j@63 92B!;6-9!"-!2"44"-A!<02"4"0-B!.-9!0-!: 6/j 0 10 20 30 40 Time (ms) -80 -60 -40 -20 0 Degrees (¡) Subplot Roll 0 10 20 30 40 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 0 10 20 30 40 Time (ms) 250 252 254 256 258 260 Estimated Altitude (m) Subplot Altitude 0 10 20 30 40 Time (ms) 0 2 4 6 8 10 Acceleration (g) Subplot RMS Figure 93 Example of resulting roll, pitch, alti tude, and RMS values from subject 18 ?6;;! : 6/j@63 92B!;6-9!"-!2"44"-A!<02"4"0-B!.-9!0-!: 6/j

PAGE 161

! $(' Q Fall Backwards Graphs 240 250 260 270 280 Time (ms) -100 -50 0 50 Degrees (¡) Subplot Roll 240 250 260 270 280 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 240 250 260 270 280 Time (ms) 1570 1580 1590 1600 1610 Estimated Altitude (m) Subplot Altitude 240 250 260 270 280 Time (ms) 0 2 4 6 8 Acceleration (g) Subplot RMS Figure 94 Example of resulting roll, pitch, altitude, and RMS values from subject 9 ?6;;"-A!:6/j@6392 330 335 340 345 350 355 Time (ms) -100 -80 -60 -40 -20 0 20 Degrees (¡) Subplot Roll 330 335 340 345 350 355 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 330 335 340 345 350 355 Time (ms) 1580 1585 1590 1595 1600 Estimated Altitude (m) Subplot Altitude 330 335 340 345 350 355 Time (ms) 0 2 4 6 8 Acceleration (g) Subplot RMS Figure 95 Example of resul ting roll, pitch, altitude, and RMS values from subject 10 ?6;;"-A!:6/j@6392

PAGE 162

! $(( 260 270 280 290 Time (ms) -100 -80 -60 -40 -20 0 20 Degrees (¡) Subplot Roll 260 270 280 290 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 260 270 280 290 Time (ms) 1400 1450 1500 1550 1600 1650 1700 Estimated Altitude (m) Subplot Altitude 260 270 280 290 Time (ms) 0 5 10 15 20 Acceleration (g) Subplot RMS Figure 96 Example of resulting roll, pitch, altitude, and RMS values from subject 11 ?6;;"-A!:6/j@6392 285 290 295 300 305 Time (ms) -100 -50 0 50 Degrees (¡) Subplot Roll 285 290 295 300 305 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 285 290 295 300 305 Time (ms) 1450 1500 1550 1600 Estimated Altitude (m) Subplot Altitude 285 290 295 300 305 Time (ms) 0 2 4 6 8 10 Acceleration (g) Subplot RMS Figure 97 Example of resulting roll, pitch, altitude, and RMS val ues from subject 12 ?6;;"-A!:6/j@6392

PAGE 163

! $() 325 330 335 340 345 350 Time (ms) -100 -50 0 50 Degrees (¡) Subplot Roll 325 330 335 340 345 350 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 325 330 335 340 345 350 Time (ms) 1565 1570 1575 1580 1585 1590 1595 Estimated Altitude (m) Subplot Altitude 325 330 335 340 345 350 Time (ms) 0 2 4 6 8 Acceleration (g) Subplot RMS Figure 98 Example of resulting roll, pitch, altitude, and RMS values from subject 13 ?6;;"-A!:6/j@6392 310 315 320 325 330 Time (ms) -80 -60 -40 -20 0 20 Degrees (¡) Subplot Roll 310 315 320 325 330 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 310 315 320 325 330 Time (ms) 195 200 205 210 215 Estimated Altitude (m) Subplot Altitude 310 315 320 325 330 Time (ms) 0 1 2 3 4 Acceleration (g) Subplot RMS Figure 99 Example of resulting roll, pitch, altitude, and RMS values from subject 14 ?6;;"-A!:6/j@6392

PAGE 164

! $(* 290 295 300 305 Time (ms) -100 -50 0 50 100 Degrees (¡) Subplot Roll 290 295 300 305 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 290 295 300 305 Time (ms) 230 235 240 245 250 255 260 Estimated Altitude (m) Subplot Altitude 290 295 300 305 Time (ms) 0 1 2 3 4 5 6 Acceleration (g) Subplot RMS Figure 100 Example of resulting roll, pitch, altitude, and RMS values from subject 16 ?6;;"-A!:6/j@6392 220 230 240 250 Time (ms) -100 -80 -60 -40 -20 0 20 Degrees (¡) Subplot Roll 220 230 240 250 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 220 230 240 250 Time (ms) 200 220 240 260 280 Estimated Altitude (m) Subplot Altitude 220 230 240 250 Time (ms) 0 2 4 6 8 10 Acceleration (g) Subplot RMS Figure 101 Example of resulting roll, pitch, altitude, and RMS values from subject 17 ?6;;"-A!:6/j@6392

PAGE 165

! $(+ 150 160 170 180 190 Time (ms) -80 -60 -40 -20 0 20 40 Degrees (¡) Subplot Roll 150 160 170 180 190 Time (ms) -200 -100 0 100 200 Degrees (¡) Subplot Pitch 150 160 170 180 190 Time (ms) 150 200 250 300 350 Estimated Altitude (m) Subplot Altitude 150 160 170 180 190 Time (ms) 0 2 4 6 8 10 12 Acceleration (g) Subplot RMS Figure 102 Example of resulting ro ll, pitch, altitude, and RMS values from subject 18 ?6;;"-A!:6/j@6392

PAGE 166

! $(G R System Usability Survey

PAGE 167

! $)H "#$%&'!($)*+,+%#!"-),& Digital Equipment Corporation, 1986. Strongly Strongly disagree agree 1. I think that I would like to use this system frequently 2. I found the system unnecessarily complex 3. I thought the system was easy to use 4. I think that I would need the sup port of a technical person to be able to use this system 5. I found the various functions in this system were well integrated 6. I thought there was too much inconsistency in this system 7. I would imagine that most people would lea rn to use this system very quickly 8. I found the system very cumbersome to use 9. I felt very confident using the system 10. I needed to learn a lot of things before I could get going with this system ./0 T6/j.-k".!Q53"24.-2.! 123/45 60! $) q %H+* 7&8$+9: 0! $) q X./.1:.3 q %H$)