Docstoc

FALL DETECTION

Document Sample
FALL DETECTION Powered By Docstoc
					                                        Final Report




                              FALL DETECTION




                              ECE4007 Senior Design Project



                              Section L05, Fall Detection Team

             Nicholas Chan, Group Leader (channy@gatech.edu, 404.642.9476)

                    Akshay Patel (akshayp@gatech.edu, 404.610.7649)

              Abhishek Chandrasekhar (abhishek@gatech.edu, 864.346.8428)

                 Hahnming Lee (hahnming.lee@gatech.edu, 908.392.3876)




                                         Submitted



                                       April 28, 2009




Fall Detection (ECE4007L05)                                                  1
                                                        Table of Contents
EXECUTIVE SUMMARY ............................................................................................................ 4


INTRODUCTION .......................................................................................................................... 5


   Objective ..................................................................................................................................... 5


   Motivation................................................................................................................................... 5


   Background ................................................................................................................................. 6


PROJECT DESCRIPTION AND GOALS..................................................................................... 6


TECHNICAL SPECIFICATION ................................................................................................. 10


DESIGN APPROACH AND DETAILS ...................................................................................... 12


   Design Approach ...................................................................................................................... 12


   Codes and Standards ................................................................................................................. 20


   Constraints, Alternatives, and Tradeoffs .................................................................................. 21


SCHEDULE, TASKS, AND MILESTONES .............................................................................. 23


PROJECT DEMONSTRATION .................................................................................................. 24


MARKETING AND COST ANALYSIS..................................................................................... 24


   Marketing Analysis................................................................................................................... 24


   Cost Analysis ............................................................................................................................ 25


SUMMARY AND CONCLUSIONS ........................................................................................... 27


REFERENCES ............................................................................................................................. 28


APPENDIX A............................................................................................................................... 30


GANTT CHART FOR FALL DETECTION ............................................................................... 30


APPENDIX B ............................................................................................................................... 32


CODE ALGORITHIM AND CODE RESPONSIBILITIES........................................................ 32


APPENDIX C ............................................................................................................................... 34


Fall Detection (ECE4007L05)                                                                                                                       2
MATLAB CODE FOR FALL DETECTION............................................................................... 34


APPENDIX D............................................................................................................................... 35


WEBSITE CODE FOR FALL DETECTION ALERT ................................................................ 35





Fall Detection (ECE4007L05)                                                                                                               3
                                     EXECUTIVE SUMMARY


        Hospitals and nursing homes are experiencing a simultaneous increase in injuries due to

falls and a decrease in qualified staff hires. More than 16,000 elderly patients die because of

injuries sustained from falls each year and the severity of the issue is exacerbated by the staff-to-

patient ratio [1]. Solving the problem involves implementing preventative measures that will

minimize incidents leading to injury without necessitating a larger staff. The project used in

conjunction with a current product that aims to prevent falls from ever occurring, like the

SleepSafe bed, can help decrease many of the problems [2].

        During hours when there is a smaller on-call staff, such as at night, the patient and the

personnel have the option of activating the system to detect falls. Two cameras are strategically

placed, depending on the patient’s privacy concerns and desired level of safety, to record

movement. A feed of the video is continuously transmitted to a local computer on the network,

where the data is analyzed and processed to determine if a fall has occurred and whether it

medical assistance should be sent. The computer can differentiate between different types of

movements by looking at certain conditions and if the analyzed activity fulfills the criteria. If

help is necessary, an alarm is sent to a nurse’s station in order to alert staff.

        The project was successful in analyzing videos of prerecorded falls, but had difficulties

with live video. This was attributed to the limitations of the language we wrote the program in

and the low-end computer resources, a requirement to keep the cost down. The projected quote

came to a maximum of $500 per room, with additional rooms driving down the costs further.

        Many of the problems with the current system, primarily its slow speed and its poor

management of resources, would likely be solved if the current build was exported to a different

computer language like C++ or OpenCV.

Fall Detection (ECE4007L05)                                                                         4
                                        FALL DETECTION

        The document details the procedure for detecting potential falls by elderly people

occurring in hospitals, nursing homes, and other assisted living locations. A system with two

web cameras connected to a computer running MATLAB is set up to analyze the motion inside

the room and alert a local nurse’s station as soon as a potential fall has happened so help can be

sent, if necessary.

                                         INTRODUCTION

                                               Objective

        The task of the project is to detect falls and to alert the staff if one has occurred, leading

to a decrease in injuries occurring from falls. Falls are one of the most common injuries in

hospitals, being 40% more likely to occur in a hospital or nursing home than in other industries

and locations [3]. The elderly are approximately 50% more likely to fall than the general

population [4]. The product will appeal to bigger hospitals and nursing homes that have a very

high patient-to-doctor ratio. A requirement of the system is for the locations to have modern

infrastructure to support technology such as Ethernet.

                                              Motivation

        Falls cause 90% of the 300,000 hip fractures which occur each year [5]. The project

solves a persistent problem that has had multiple possible solutions, but none have been optimal.

Some, like a physical alarm, would be intrusive on the surrounding patients. Others, like a 24-

hour call center, would be more expensive [6]. Incorporating elements from various products in

the market today yields the best system for hospitals and nursing homes. It also aims to be one of

the lower-cost solutions.




Fall Detection (ECE4007L05)                                                                              5
                                            Background

       Previous products include iLife, a product using an accelerometer to measure falls [7].

The high number of false-positives associated with the technology as well as the requirement of

the user constantly wearing an extra piece of equipment to enable the product to work made it an

unattractive way to solve the problem. RFID solutions were initially another possibility, but the

cost significantly outweighed any of the benefits. Furthermore, hospital rooms have different

types of equipment which could cause EMI interference. Previous solutions have also used

optical flow, or edge detection, to try and prevent falls. Unfortunately, shadows caused the

algorithms to act in unpredictable ways, making it a one-dimensional product (it would only

work at night or in the absence of shadows). Lastly, pressure sensitive mats were another

potential technology to use, but presented several problems. It would require either the whole

floor to be covered, a heavy expense and difficult to clean, or it would have edges which could

cause falls, negating any possible positives from the mat being in place.



                           PROJECT DESCRIPTION AND GOALS

   •   Correctly detect a fall

   The two cameras which are connected through USB to a computer are set up to analyze the

patient and his or her movements during the most vulnerable times. When unexpected motion

that is characterized by high speed and changes in orientation occurs, the computer cross-checks

all the variables with each other and determines whether a fall has occurred. If the analysis

indicates the patient needs assistance, an alarm will be sent to the nurse’s station.

   During initial preparation, 50 videos were taken of different falls that could happen in a

hospital room and analyzed using the program. In this state, the algorithm was correctly able to


Fall Detection (ECE4007L05)                                                                         6
detect a fall and differentiate it from a non-fall with almost 80% accuracy. When it was

implemented to detect in real-time, the algorithm struggled to loop back and continuously

analyze the surrounding environment due to low computer resources and the nature of the

language the program was written in.

    While the problem presented was a major hindrance on the progression of the project, the

solution is very easily implemented and would be included in a final product sold to hospitals.

MATLAB, a simple programming language with wide variety of preset tools, was chosen at the

beginning of the development cycle. If a lower-level programming language like C++ was

chosen, the program would likely have better memory allocation and would therefore be more

likely to run in real-time.

    •   Send an alert and create a database from it

    After a fall is potentially detected, an alarm was sent to the nurse. A nurse’s station with

Ethernet access and a web browser would receive a text prompt and a flashing screen displaying

a blurry picture of the fall and the image which indicated to the system that a fall has occurred in

the indicated room. The nurse will visually confirm the analysis and assist if necessary or

indicate that it is a false-positive.

    At the time of the proposal, a specific method of the alarm system had yet to be chosen. The

possibility of a hardware based solution was explored and a circuit was built which could beep,

but it required a serial connection, an extra and unnecessary cost. The website method of

displaying rooms’ statuses with a blurred picture was chosen because there would be little cost

and any computer with network connectivity would be able to display the relevant information.

    Unfortunately, the database was not built due to both logistical reasons and computer

resourcing. Because of the high volume of patients in one room and the differing nature of how a



Fall Detection (ECE4007L05)                                                                        7
person falls, a database would likely confuse the system and force it to think there were more

false-positives. The best way for the system to learn from its past falls is to determine the

thresholds and constantly update it, a feature that could easily be implemented in the final

product.

   •   Cost around $500 to implement

   Cameras bought at retail cost $69.99, but can either be downgraded or bought in bulk to

reduce cost. The specific model used in the project is a Microsoft VX-6000. A computer with

MATLAB installed can be bought or built for under $300 and would be able to support a

network of rooms. If the infrastructure is already in place (a network for internet), the cost to

wire rooms would be minimal and would not require significant manpower or time. The cost

would be a one-time flat rate, excluding any maintenance, which would be minimal and could be

performed offsite.

   The goal was to create a room that cost as little as $100. Unfortunately, the cost of MATLAB

and the high resources forced the system to have one computer per room as opposed to one

computer running multiple rooms. Furthermore, two cameras had to be used to increase the

reliability of the program. A maximum cost of $500 was calculated if only one room was

implemented at a time, and further resources could be dedicated to allowing one computer to

analyze multiple rooms with an extended development cycle.

   •   Accommodate the privacy of the patient and his or her family

   In the project, an environment is designed and optimized to help a specific patient and his or

her needs. A camera is placed to capture as much of the area of the room as possible, or, if the

patient opts to choose an option to better protect his or her privacy, placed to minimize fears of

being constantly monitored while still detecting falls.



Fall Detection (ECE4007L05)                                                                          8
    A consent form would be signed by the patient and his or her family upon entry into the

nursing home or the hospital. It would only ask patients in the target demographics, like the

elderly, or those whom required the assistance of others to move. If the patient refuses or only

wants limited coverage, the cameras could be taken away, turned off, or moved. The patient

would also have an option of turning it on or off in certain situations so as to not be monitored at

all times. This would both help with computer resources as well as accommodate many of the

requests concerning the cameras.

    When a fall is detected, a blurred image would be sent to the nurse’s station as opposed to a

regular image. The only information contained would be the patient room. The image would be

blurred enough to be able to display a clear image of a possible fall while also protecting the

identity of the patient in question.

    The sensitive issue, while not originally included in the proposal, became a bigger concern as

the project progressed and more research was done concerning liability.

    •   Appeal to larger hospitals and nursing homes

    The product will be marketed towards middle-tiered hospitals and nursing homes that are

under-staffed and overpopulated. It will help them reduce costs while simultaneously preventing

injuries occurring from falls. It is not marketed towards younger people and customers staying at

home without assistance. The algorithm is optimized towards the elderly, as hey are more likely

to be unable to seek help.

    Testing done proved that movement of the younger will be more likely to cause unlikely

scenarios and those that are currently unaccounted for in the program.




Fall Detection (ECE4007L05)                                                                         9
                               TECHNICAL SPECIFICATION

  Table 1. Microsoft VX-6000 Web Camera Specifications

     Video Resolution          Frame Rate           Audio Support             Computer
                                                                              Interface
      160x120 (scaled            15 FPS                    No                   USB
     from 1280x1024)


  Table 1 displays the relevant specifications for the camera. The default settings included a

relatively large 1280x1024 resolution and a speed of 30 frames per second (FPS). To decrease

the demand on the computer’s resources and because of the lack of a need for larger, more

detailed images, the settings were adjusted and scaled down. The camera can be set in a fixed

position, meaning viewing angle is irrelevant. The frame rate at 15 FPS will be able to capture

any fall quickly enough. The size could be much higher, but testing showed that small feeds will

still capture the image and identify the person and the motion in the room. The computer is

connected through a standard USB connection.

  Table 2. Dell Inspiron 530 with Windows Vista Home and MATLAB 2008b

       Processor                                      2.50GHz Dual Core Processor
       USB ports                                      3
       Memory                                         3072 MB of DDR3 Ram
       Line Voltage                                   100V to 240V AC
       Frequency                                      50 to 60 Hz

       Table 2 displays the test server being used. The computer will only be connected to the

camera and a power outlet. Since the webcam will be running and streaming Windows

proprietary media format WMV, a machine running Windows Vista Home with 3072 MB of

RAM will handle the feed. If a different model web camera was to be used in a different system,

any operating system capable of running a version of MATLAB could be used.




Fall Detection (ECE4007L05)                                                                       10
        During testing, a variety of computers with different technical specifications were used to

test the efficacy of the program across different builds.

   Table 3. Sony Vaio with Windows XP and MATLAB 2008b

        Processor                                      1.8GHz Core Solo Processor
        USB ports                                      2
        Memory                                         1540 MB of DDR2 Ram
        Line Voltage                                   100V to 240V AC
        Frequency                                      50 to 60 Hz


        The low-end of the computers is shown in Table 3.

   Table 4. Alienware Area-51 m15x Laptop with Windows Vista and MATLAB 2008b

        Processor                                      2.00GHz Core 2 Duo
        Graphics Card                                  256 MB NVIDIA GeForce 8600M GT
        USB ports                                      3
        Memory                                         2048 MB of DDR2 Ram
        Line Voltage                                   100V to 240V AC
        Frequency                                      50 to 60 Hz


        The other machine specification is shown in Table 4. The Alienware took, on average, 13

seconds to run the algorithm. The VAIO took around 18 seconds to run the same video and

analysis. The large disparity between the experimental times was evidence that superior

hardware would produce superior results. The two most used components during analysis of the

program were RAM memory and the processor. Video cards and hard drives were extras and a

cost that could be controlled when building the final machine.

        The decided specifications were used because it fulfilled the main two requirements,

processing power and memory size, without exceeding the desired costs. Since it is

manufactured as a workstation, buying multiple computers could drive down the cost of buying

it directly from Dell, the supplier.



Fall Detection (ECE4007L05)                                                                     11
                            DESIGN APPROACH AND DETAILS

                                          Design Details

       While the goal is to prevent falls from occurring, the problem can at least be mitigated

with the prevention of many injuries from falls. The project creates a set-up with two cameras

and one computer able to provide immediate assistance through an alert being sent to a nurse at a

nearby station connected through an Ethernet network.

       Two USB cameras, specifically the VX-6000, will be connected to a computer using

standard USB 2.0 cables. Patients will have two options for how the camera could be set up.




                 Figure 1. Camera setup allowing for maximum coverage.



       Figure 1 shows how the cameras would be set up to cover the most area in the room. It

would be able to capture the whole area where a fall could potentially occur. All uncovered areas

are either inaccessible by the patient or else covered with another piece of furniture.




Fall Detection (ECE4007L05)                                                                       12
                 Figure 2. Camera setup allowing for maximum privacy.



        Figure 2 shows the camera setup which would allow for maximum privacy while still

performing the product’s primary purpose of detecting falls. The second camera would be placed

at knee level so as still record the movement of the person, but obstruct the sight of the camera

from the patient. It could still correctly detect falls.

        The computer could be located in whatever location would be the least obtrusive to the

patient or at another location still in proximity to the room.

        The cameras, as stated in the technical specifications, capture video at 15 FPS. The feed

is sent to the computer and analyzed in MATLAB with the aid of the image acquisition toolbox,

an extra bought with the MATLAB license. An established connection has been tested and

MATLAB is a capable program of rendering the real time feed and analyzing it. The lower FPS

would decrease the workload of the processor as less data would need to be processed, increasing

the efficiency of the program.

        Once processed into MATLAB, the written program will analyze the image to detect

whether or not the motion constitutes a fall. The theoretical outline is diagramed in Appendix A.
Fall Detection (ECE4007L05)                                                                         13
       Once the system is activated in the room, foreground segmentation takes place. The

process involves isolating the person from the background. Since the room may have other parts

of it moving at different times, the function would then assign each part either as the background

or the foreground. The background will appear as black and the foreground as white.




              Figure 3. Foreground segmentation in action.

       Figure 3 demonstrates how the function creates a binary foreground image. A clear

outline of the test subject can be seen as well as spots around it labeled as the foreground. To

negate the shadow effect that inevitably occurs with each image, a HSV (hue, saturation, value)

color space is used.

       After this process, the person must be isolated from the rest of the pixels which are

extraneous for the processing algorithm. Another function is used which will identify the largest

“blob” in the picture and designate the rest of the foreground pixels as background. This

eliminates all noise and decreases the possibility of false-positives occurring from motion in the

room not involving the person.




Fall Detection (ECE4007L05)                                                                        14
              Figure 5. The blob detection eliminating the rest of the noise.


       Figure 5 demonstrates the elimination of noise after the blob detection function is

performed.

       After this, the video feed will analyze the images frame-by-frame and use its history to

detect if a fall has occurred. This is called Motion History Imaging (MHI). MHI helps identify

and analyze variables relating to a fall like orientation and the velocity of the motion. The

velocity is referred to as the motion coefficient. The value is between 0 and 1, with 1 being

extreme motion and 0 being none.




             Figure 6. Motion coefficient and its different states.


Fall Detection (ECE4007L05)                                                                       15
        Figure 6 displays how the program evaluates the motion coefficient when the video is

analyzed. The gray pixels in each image represent movement while white pixels represent the

person in his or her current state. The larger the number of gray pixels in proportion to the

number of white pixel, the more likely a fall has occurred.

        While this variable will help recognize falls, it is a necessary but not sufficient condition

to identifying a fall. To increase accuracy, several factors are measured. The next variables relate

to elliptical approximations.

        Once the blob is identified, an ellipse is also drawn around the person and changes as the

person moves. An ellipse was chosen because it has multiple characteristics that are helpful

towards determining a fall. The directions of the major and minor axis determine the ellipse’s

orientation, while the ratio between the two axes determines the eccentricity. Sudden changes in

either are strong indicators of a fall.




                Figure 7. Elliptical approximations from one frame to another.




        Figure 7 shows the difference between the ellipse during normal motion and a fall. Not

only is the ellipse a different shape, but the orientation has changed significantly. Both factors




Fall Detection (ECE4007L05)                                                                          16
independently may indicate a fall, but used in conjunction with the motion coefficient, provide

an even more accurate basis for fall detection.




                Figure 8. MATLAB analysis of angle and eccentricity.




       Figure 8 shows additional analysis done by MATLAB and the typical plot of angle and

eccentricity.

       After testing, the two best indicators of a fall were the motion coefficient and the standard

deviation of the angle from frame-to-frame. Testing indicated that a motion coefficient of .65 or

higher with a standard deviation of the ellipse angle of .60 radians or higher indicated a fall. This

was highly dependent on camera position, but all of the tests done were done with the same

camera position, meaning a similar environment could be replicated. Factors like the auto-

adjustments in the lighting of the camera as well as fall motions of different subjects somewhat

skewed results, but with these tested thresholds, the results were nearly 80% accurate.




Fall Detection (ECE4007L05)                                                                       17
  Table 5. Results of the falls and non-falls.

                     Category               % Success                 % Fail

                        Falls                 83.33%                  16.67%

                     Non-falls                   75%                   25%



       Table 5 shows the results of the testing using the thresholds in place. Falls were detected

from 50 videos, 30 actual falls and 20 non-falls.

       Once a fall is detected, an alarm is set off. A message is sent to a website at a nurse’s

station running in the background. The website refreshes every 5 seconds automatically.




           Figure 9. Website status when no falls have occurred.


       Figure 9 displays a screen shot of how the monitored website would look when no falls

have occurred. Each website can be built for a hospital depending on its needs. The website is

intentionally simple as it can run on multiple machines without hogging large amounts of the

computer’s resources.



Fall Detection (ECE4007L05)                                                                        18
       When a fall occurs, a blurry image is sent to the website and it is updated, alerting the

nurse through a flashing background at the station and a text alert.




            Figure 10. Website status as soon as a fall occurs.




       Figure 10 shows that the room 1 picture is updated to reflect that a fall has potentially

occurred. As soon as a nurse sees this, he or she can determine if an actual fall has occurred and

go help if it has. After assistance is doned, the nurse has the option of archiving the fall and

resetting the website to its normal state again.




Fall Detection (ECE4007L05)                                                                        19
        Figure 11. The option for a nurse to archive the fall.


       Figure 11 shows the website interface for either archiving or reloading the page after a

fall has occurred and having the option of marking it as a fall or as a false positive to be ignored.


                                       Codes and Standards

The codes and standards requiring adherence in the project are below.

   •   Ethernet

       How the computers are connected to each other through a network.

   •   Universal Serial Bus (USB)

       The computer and the cameras will communicate through a USB connection.

   •   National Health Services (NHS)

       The network would have to be deemed safe to be built into a hospital.




Fall Detection (ECE4007L05)                                                                        20
   •   National Electric Code (NEC)

       The standard for safe installation and electrical wiring.

       A confidentiality agreement was used in order to set the guidelines between the patient

and the hospital. To comply with the NHS confidentially code of practice, the images sent to

alert the nurses at the stations will be blurred and will only include the number of the room.

Personal information relating to the fall will not be saved in the system.


                             Constraints, Alternatives, and Tradeoffs

       As mentioned earlier in the report, there are several other alternatives to the product

proposing to do the same task.

       Pressure sensitive mats could have been used and implemented in each room. This would

have recorded movements by the weight at certain points. If a large mass was distributed in a

certain way on the mat, a fall would have been detected. The main disadvantage to this product

was that it could possibly cause even more falls because of the necessary edges that come from

placing a mat over the ground. The large area needed to place it on a hospital room would also

drive up costs. The mat would likely also require special cleaning and more maintenance than

most floors do in modern hospitals. Many of the costs built into the current build of the project,

like the computer and the website development for alerting the nurses, would still be built into a

project centering on pressure sensitive mats, making it prohibitively more expensive.

       An RFID solution was also a possibility. The cost of implementing it made it one of the

least desired choices. Installing a RFID receiver can cost upwards of $1000 for a receiver and the

EMI interference it could potentially create in a hospital room with many different types of

machines [8].



Fall Detection (ECE4007L05)                                                                      21
       Optical flow is a technique using edge detection for falls. While the cost would be

minimal, the inherent method had difficulty differentiating between shadows and the person,

causing inaccurate analysis. This could not be solved with higher costs or superior equipment

and would likely require an even longer development cycle.

       Lastly, accelerometers were another potential path, but were seen as too unreliable and

antiquated to use effectively. Forcing patients to wear one could be invasive and cause unknown

effects, especially if the patient can take it off in vulnerable times, like when or she is asleep.

       Privacy was a major concern and restricted some of the possible features. The system

would never store any video of the actual fall occurring and would also never store any image

without it being blurred. The computer would have no information about the patient except the

room number and location. Furthermore, the only people accessing the information about the

falls would be personnel like nurses and doctors, who would also have to sign forms in order to

see the images. Forcing entering patients to sign a terms-and-conditions agreement specifying

the level of monitoring would also be required to minimize litigation and liability [9]. The

privacy issue, as well as some logistical issues, eventually led to the elimination of the database

feature. This did not directly hinder the development of the project, as most of the analysis would

happen in real-time in independent instances.

       The other two constraints in this project are directly linked to the amount of money

willing to be spent, which is often at the discretion of the customer. The algorithm and program

do not gain efficiency when a higher resolution or faster camera is used, meaning the cost of

cameras can be kept on the lower end. The main constraint is the power of the processing of the

computer. A workstation with a large amount of memory and a fast processor may be able to

analyze and send a signal quicker than one without the high-end components. The cost would be



Fall Detection (ECE4007L05)                                                                           22
most applicable if a customer were to use one computer to analyze multiple rooms, possibly

slowing down response time and decreasing the overall quality of the product.

       Also, the camera is compensating for its small lens and features by using auto-lighting for

different situations. This is useful to normal users, but actually causes problems when doing

analysis because it can cause differences in the motion and change many of the

foreground/background pixels. The tradeoff for using a lower cost camera was to deal with this,

which was done by setting the contrast high enough where the light would not change. The

actual image clarity was not as important as the detection of the person, meaning the contrast

being high would not affect the final analysis.

                         SCHEDULE, TASKS, AND MILESTONES

       A Gantt chart is included in Appendix B. Since it is particularly large, a complete chart is

accessible on the group’s web site.

       Because the hardware of the proposed project is both simple to set-up and acquire, most

of the work was done on the software level. The modular nature of the project allowed group

members to split the software development into parts which can be concurrently developed by

different people. This not only makes the project easier to write and compile, but also helps hold

others accountable, as a missing function would prevent the whole project from working.

       The difficulty is mainly placed in the writing and optimization of the program. The

specific tasks are assigned to each group member (as dictated in Appendix A).

       The project went according to plan. The only part not directly incorporated into the Gantt

chart that was worked on later was the method for implementing the alarm. Any time a group

member did not have a specific task, one was assigned to research and work on a possible




Fall Detection (ECE4007L05)                                                                      23
method until it was decided that the web site should be used. Otherwise, the work done on the

project was in line with the initial projection at the beginning of the semester.

       Each group member was involved with different parts of the project, including editing

and writing the reports as well as software, and tasks were divided evenly.

                                 PROJECT DEMONSTRATION

       Due to the limitations described in the previous parts of the report, there was no live

demo performed. Tests were done with prerecorded videos. Initial plans to demo the actual

camera detection system were eliminated because of the lack of proper equipment as well as

flaws in the current build of the project.

       The test of prerecorded video, using the Alienware machine described in Table 4, was

correctly able to differentiate between falls and non-falls by checking the thresholds and the

multiple variables measured during movement. The nearly 80% success rate is relatively high for

the program as it could be further refined and the limitations could also be attributed to the

equipment used.

                            MARKETING AND COST ANALYSIS

                                         Marketing Analysis

       Current market analysis is aimed towards home consumers and others. Items requiring

24-hour call centers, like the Brick House Alert, are unsuitable for hospitals because patients are

already in the hospital with constant supervision [10]. The product poses an unnecessary monthly

cost on them and would eliminate any benefit as help is already present at the hospital. A

physical and loud alarm, like the Patient Alarm & Fall Down Safety Alert offered at the Survival

Store online, is unsuitable for hospitals because it will disturb other patients [11]. Other products

have also used a sensor attached to a part of the body to analyze if a sudden motion has occurred,


Fall Detection (ECE4007L05)                                                                        24
but it is undesirable because it creates more false alarms and would require different analysis for

different body types. These, while fulfilling certain segments, fail to prevent injuries. The system

built has lower false-positives. It is able to notify help without causing unnecessary panic. It is

significantly cheaper, because the product will have a one-time cost as opposed to a monthly cost

(like in a 24-hour call center).


                                            Cost Analysis

        The cost of the product is relatively low, presenting yet another advantage to most

customers. The cost of parts is listed below.

    •   Two Microsoft VX-6000 cameras – $69.99 each

    •   Dell Inspiron 530 - $270.00 each

    •   MATLAB license - $4750.00 one time fee

    Each price listed above is the individual price on the website and the retail price. Each would

be significantly lower as the demand for the product increases and more parts are needed to be

bought in bulk. For example, while the retail price for the camera is $69.99 directly from

Microsoft, the street price on it is as low as $30.00.

    The cost for labor is listed below.

        •   Estimated installation costs - $70.00

        •   Development price - $45,000.00

        The installation cost would be a one-time cost applied to each room. It is also estimated

with the assumption that an Ethernet network is already in place. The product is only aimed at

places with the necessary infrastructure already in place, like more modern hospitals. The

development cost was calculated when charging four engineers $55.00 per hour during a 10-

week development cycle while working 20 hours a week between all of the group members.

Fall Detection (ECE4007L05)                                                                           25
       The only cost that should have the option to be upgraded is the computer. If a computer is

already in place that could serve the same function, the cost of the computer would be subtracted

from the initial quote to the hospital. Furthermore, the cost can go even further down if a

mainframe server is built into the hospital that could serve this function. This feature could be

included with more development.

       If a cost of $500 dollars per room was charged initially, the group would be losing

money. There are three factors necessary to turn a profit. The first would be for computer

technology to advance in a similar fashion as it has in the past, driving down costs for parts. It is

unnecessary to upgrade further than what is already in place, meaning in over a year, the

difference between current cost and future project price would turn into profit. The second would

be increased demand. As more orders come in, the parts of each room will cost less and less as

they are bought in large quantities and the differences between the quoted price and the cost in

parts would turn into profit. Lastly, little to no maintenance is required and further development

can be done on request. This would mean that the costs of further developments would be

minimal and could be charged on a client basis. In the first year, the product would be making no

profit and would be losing money as the group would be working for nearly nothing. After the

first year and assuming some sort of growth, the estimate is that a company could have a 20%

profit from each unit sold. Assuming all of these factors, it is estimated that within five years, a

company could turn a significant profit and expand the scope of the project.

       As the product has been created now, it could not be sold to most modern hospitals

because of the limitations of dealing with real-time. Additional development would be required

in order to fully export and test the code in C++. An estimated three more weeks would be

required for the product to function at any hospital. No additional software or hardware is



Fall Detection (ECE4007L05)                                                                         26
necessary unless the hospital requires a more advanced computer. The computer suggested in the

report serves as a baseline and the minimum requirements for the program to function in its

intended manner.

                               SUMMARY AND CONCLUSIONS

        The project, in its current state and with the projected development, has achieved the

goals set from the onset. It is successfully able to detect falls and alert a nurse’s station for help

in an appropriate amount of time. The cost of the project was relatively low and allowed for large

flexibility for future development.

        The critical mistake made was choosing MATLAB as the development language. It was

chosen because of its supposed ease of use and built-in tools, but many of its simplifications

caused the most problems. This mistake did not cause the project to be a failure, as it can be

easily transitioned into a more accessible language like C++ or C.

        Also, projecting a database to be used later on was short sighted as it assumed that people

would fall in the same way and could be aggregated together. With individual ways people fall

and the different body parts, the database proved to be an impossible goal that would actually

hinder the development of the project. It would also cause numerous privacy concerns.

        The project, with minor adjustments, could be marketed and sold. Future development

would include optimizing code, but this is a goal that would not have a time table as it could be

pursued for an extended period of time. The code will never be as fast as it can be and more

development can always be done. If the product could be packaged and sold with a product that

tries to actively stop falls, it could create a more profitable model and one that addresses both

problems concerning falls in hospitals and nursing homes.




Fall Detection (ECE4007L05)                                                                          27
                                        REFERENCES

[1] (2009, Jan. 19). “Falls Among Older Adults: An Overview,” [Center for Disease

       Control and Prevention], [cited 2009 Apr. 28], Available HTTP:

       http://www.cdc.gov/HomeandRecreationalSafety/Falls/adultfalls.html

[2] J. Tombetti (2008, June 28). “SleepSafe Beds – Not Your Typical Hospital Bed,” [DOTmed

       News], [cited 2009 Apr. 28], Available HTTP: http://www.dotmed.com/news/story/6243/

[3] “Simple strategies can reduce falls and liability: women and elderly fall more frequently,”

       Rehab Continuum Report, Nov. 2004.

[4] “Oldies, depressed people more likely to take a tumble,” Thaindian News, 18 June 2008.

[5] (2007, Aug.). “Your Orthopaedic Connection: Prevention of Falls: Facts,” [American

       Academy of Orthopaedic Surgeons], [cited 2009 Apr 28], Available HTTP:

       http://orthoinfo.aaos.org/topic.cfm?topic=A00101

[6] S. Lord, C. Sharrington, and H. Menz, “Epidemiology of Falls and fall-related injuries,” in

       FALLS in older people: Risk Factors and strategies for prevention, 1st ed. Cambridge,

       England: Cambridge Univ. Press, 2001, ch. 1, pp. 3-13.


[7] “iLife Solutions, Inc.,” [iLife Solutions, Inc.]. [cited 2009 Apr. 28] Available HTTP:

       http://www.ilifesolutions.com/products.html

[8] (2008, Aug. 1). RFID EMI poses medical equipment hazard. Health Management

       Technology [magazine]. Available: http://www.allbusiness.com/medicine-

       health/diseases-disorders-cardiovascular-disease/11602936-1.html

[9] D.L. Gray-Miceli, “A Nursing Guide to the Prevention and Management of Falls in

       Geriatric Patients in Long-term Care Settings,” Medscape Today, 19 May 2005.

[10] “Fall Detection,” [Brick House Alert], [cited 2009 Jan 21], Available

Fall Detection (ECE4007L05)                                                                       28
       HTTP:http://www.brickhousealert.com/howitworks.html

[11] “Patient Alarm & Fall Down Safety Alert,” [Survival Store], [cited 2009 Jan 21], Available

       HTTP: http://www.survivalstore.com/r6s15lbb4.html




Fall Detection (ECE4007L05)                                                                  29
                                APPENDIX A

                      GANTT CHART FOR FALL DETECTION




Fall Detection (ECE4007L05)                            30
Fall Detection (ECE4007L05)   31
                              APPENDIX B

               CODE ALGORITHIM AND CODE RESPONSIBILITIES




Fall Detection (ECE4007L05)                                32
Fall Detection (ECE4007L05)   33
                                APPENDIX C

                      MATLAB CODE FOR FALL DETECTION




Fall Detection (ECE4007L05)                            34
                              APPENDIX D

                 WEBSITE CODE FOR FALL DETECTION ALERT




Fall Detection (ECE4007L05)                              35
Source code and additional mentioned resources can be found in the following zip:

http://www.ece.gatech.edu/academic/courses/ece4007/09spring/ece4007l05/ak9/FallDetect.zip




Fall Detection (ECE4007L05)                                                                 36

				
DOCUMENT INFO