Sie sind auf Seite 1von 6

FALL DETECTION METHOD USING MOTION

HISTORY OF IMAGE IN THE HOME CARE


APLICATION SYSTEM
Sidiq Syamsul H.1, Rizki Aditya C.2, Agus Rochadi 3, Thomas Agung 4, Eddy Triyono 5, dan Suhendro 6
Dept. Electrical Engineering, Telecommunication Engineering, State Polytechnic of Semarang, Indonesia
cuizkye69@gmail.com

Abstract- The growth rate of the elderly in Indonesia is been developed using an accelerometer and gyroscope
increasing every year. According of the World Health sensor but there are weaknesses of the use of the fall
Organization (WHO), falls is the second largest of death in the detection device because it requires the user to install or
world, amounting to 646,000 times per year. For this reason, this
carry on the body in order to work and has a high risk of
study has designed a fall detection system using video (offline),
damage or loss is quite high [3].
using the Motion Histoy of Image (MHI) method. The tests were
carried out by inputting the video (offline), falling with a camera
Other studies in detecting falls are using smartphones.
distance of 300cm, 350cm, and 400cm, high cameras 70cm, Sensor data is taken from smart phone users placed on the
100cm and 140cm, and light intensity 310 lux and 959 lux on the left chest. When people fall, the system will detect and
Raspberry Pi 3. Fall detection using Raspberry Pi 3 shows activate the alarm system [4].
optimal results at a distance of 300cm - 350cm using the Motion History Image (MHI) is one method to detect
environment using natural light sources with a light intensity of someone's movements, the recorded video is divided into
959 lux and at a distance of 300cm with a light intensity of 310 several images, from many of these images will be
lux. The Motion History of Image method produces the
processed to find out the movement of objects that are from
C_Motion parameter value that will be used as a reference for
changes in the pixel, when the normal movement of image
fall detector. MATLAB R16 software is used to show the
visualization of falling object on the motion history of the
pixels does not experience large changes, whereas when
image.The results of this study indicate an accuracy of 82.5%, abnormal movements such as falling, the movement of
sensitivity 80%, and error 17.5% for data on falling or not objects changes from standing, sitting or sleeping, then
falling conditions. being quiet, the pixels will change very quickly [5].
There are studies relating to fall detection is to use a
Index Terms—C_Motion, Fall detection, Motion History of camera [6]. In the fall detector using a camera is a good
Image, Raspberry Pi 3, MATLAB R16 solution to be used in detecting falls because the camera
can analyze the condition of objects without touching them
I. BACKGROUND directly.

The growth rate of the elderly in Indonesia is increasing II. BASIC THEORY
every year. BPS projects that by 2045, Indonesia will have
around 63.31 million elderly people or reaching 20 percent A. Motion History of Image
of the population. In fact, the UN projection will also Motion History Image (MHI) is one method to detect
announce the percentage of elderly people in Indonesia will someone's movements, the recorded video is divided into
reach 25 percent in 2050 or around 74 million elderly [1]. several images, from many of these images will be
According to the World Health Organization (WHO) the processed to find out the movement of objects that are from
number is the biggest cause of death in the world, changes in the pixel, when the normal movement of image
amounting to 646,000 times per year. pixels does not experience large changes, while when the
Health care at home (Home care) is the provision of movement is not normal like falling, the movement of an
services and professional equipment for patients or families object changes from standing, then sitting or sleeping, then
at home to maintain health. There is technology for health being quiet, the pixels will change very quickly [5].
services in Indonesia, the Telehealt application. Telehealth This Motion History Image (MHI) was first introduced by
service systems use the internet with video, email, sms, and Bobick & Davis where pixel intensity represents the
cell phone systems [2]. novelty of movement in an image or video and in general
The design of a fall detection surveillance system has Motion History of Image is used for recognition activity[7].

1
Journal of Telecommunication, Electronic and Computer Engineering (JTEC)

To get an MHI, the first step is to extract the binary C. Confusion Matrix
sequense motion of object D (x, y, t) from the original Confusion matrix is a method used for evaluating
image I (x, y, t). classification models. Know any data classified correctly or
Then each pixel of Motion History of Image Ht is a not. One of the benefits of confusion matrix is seeing
function in the time span t, the equation is : whether the system is right or not in classifying between the
two options [9].
(1) confusion matrix which classifies predictive data
Information : comparisons with outcome data into 4 categories: True
x = pixel position at x coordinates. positive (number of positive results classified as positive),
y = pixel position in the y coordinate. false positive (number of negative results classified as
𝐻𝜏 (x, y, t) = Motion History Image (MHI). positive), false negative (number of positive results
D (x, y, t) = Binary sequence of object movements. classified as negative), true negative (the number of
negative results classified as negative). The general
The result is the value of Figure 1. where the movement equation of confusion matrix is:
of the object will turn white, the following is the result of
motion history of image (MHI).
(2)

(3)

(4)

III. SYSTEM PLANNING


Figure 1: Fall detection using MHI

B. Digital Image Representation


Image is a picture, similarity or imitation of an object.
Image as the output of a data recording system can be
digital. Digital images are images that can be processed by
a computer. Digital images are obtained using optical
sensors, after the optical sensors then digital images form a Figure 3: Research implementation method
two-dimensional plane (x, y). The intensity of the sensor
received at each point (x, y) is symbolized by f (x, y), In Figure 3. the fall detection design with motion
where x (row) and y (column) are the coordinates of the history of image requires the installation of raspbian,
position f (x, y) which states the intensity of the color level python environment and opencv as a library. Then the
of the pixel at the point [8]. test data retrieval is done by adjusting the camera
The results are in Figure 2. The coordinates of digital position and light intensity. Then an analysis is made
images are M and N, where M is a row and N is a column. and a conclusion is reached.
Then in Figure 4. explains that the design of the fall
detection method system is to use Raspberry Pi 3 model
B as a mini pc, where the object detection process is
done by getting input in the form of video recording
(offline) with a resolution of 640x480 using a digital
camera. The screen resolution of the fall detection
system on the Raspberry Pi is set at 500 x 500 pixels. it
is intended to ease the work of the mini pc. When the
system detects an object, then the system will process a
boundary box and when a fall is detected the system
will do a screenshot.

Figure 2. Digital Image Coordinate

2
Journal of Telecommunication, Electronic and Computer Engineering (JTEC)

Figure 5: Test room area


In Figure 5. A deliberate fall is taken with a light
intensity of 310lux and 959 lux. Retrieval of data is
done by observing changes in the value of the object's
falling speed based on distance and height that are
different.

V. RESULTS AND ANALYSIS


Figure 4: Flow chart system
In this chapter the researcher will collect data for later
The focus of this research is to find out the analysis according to the desired parameters. The data
performance of the fall detection system using that has been taken at the time of testing is the intensity
Raspberry Pi 3 Model B with motion history of image of the light, and the position of the camera, by falling in
and to know the level of accuracy, sensitivity, and error. a standing position, which is then examined the
In the use of fall detection the results of fall detection accuracy of the fall detection method system.
are then analyzed using a confusion matrix to determine After getting the light intensity and distance of the
the value of accuracy, sensitivity and error. camera with good results, then the test was repeated
over and over 20 times with data falling and not falling.
Then from the data taken Accuracy calculations using
IV. SCENARIO TESTING equation (2), for sensitivity calculations using equation
(3) and Error with equation (4).
Data retrieval phase in this research is to take video
data as fall fall input data using a camera that has been A. Test results at 310lux
installed inside the house. Tests related to the fall The results of testing the fall detection system work
detection method are carried out in accordance with the using 310lux light intensity using the lightmeter
parameters specified are camera positions with a camera application, carried out a trial shifting the position of
height of 70cm, 100cm, 140cm and with a camera the camera with a camera height of 70 cm, 100 cm, 140
distance of 300cm, 350cm and 400cm. As well as the cm and a distance of 300 cm, 350 cm, 400 cm.
brightness level in the testing environment in the room The results of fall detection sensitivity with 310lux
with a light intensity of 310lux and 959lux. The area of light intensity are shown in Figure 6. It can be seen that
the room used is 3 meters x 4.5 meters. the fall detection experiment with the camera position
with 310 lux light can be seen that the highest
confidence value is obtained at the height of the camera
140 cm with a distance of 300 cm that is equal to 80%
and at the height of the camera 70 cm with a distance of
350 cm that is equal to 80%, while the lowest

3
Journal of Telecommunication, Electronic and Computer Engineering (JTEC)

confidence value is obtained at the height of the camera C. Test results using software MATLAB
70 cm with a distance of 400 cm, 100 cm at a distance Tests carried out are by inserting or inputting video
of 400 cm and at a height of 140 cm with a camera recordings of people falling into the matlab software.
distance of 400 cm that is equal to 40%, due to the The video used has a resolution of 640 x 480. To
difficulty of the system in detecting falls if the distance display the movements of people falling, the parameter
between the object and the camera is too far away. used is the Motion History of Image, which is 𝐻𝜏 (x, y,
t - 1).

Figure 8: Display people falling with Motion History of


Image
Figure 6: Graph of fall detection sensitivity value with The parameter for falling person is if the pixel speed
light intensity 310lux changes (0.6). In the fall detection programming using
B. Test results at 959lux MATLAB Software, the parameter value of pixel speed
The results of testing the fall detection system work change (t) can be seen in Figure 9 [10].
using 959lux light intensity using a lightmeter Where in the picture there are objects that are
application, carried out a trial shifting the position of recognized as falling and forming a history in the frame.
the camera with a camera height of 70 cm, 100 cm, 140 The history is pixels that have speeds greater than (t>
cm and the distance of the camera 300 cm, 350 cm, 0.6).
400cm.
The results of sensitivity fall detection with 959lux
light intensity are shown in Figure 7. It can be seen that
the experiment with the position of the camera with 959
lux light can be seen that the lowest confidence value is
obtained at the height of the camera 140 cm with a
distance of 400 cm that is equal to 40%, due to the
Figure 9: Motion History of Image detection results
difficulty of the system in detecting fall when the
distance between the object and the camera is too far 1. Analysis of fall detection test with light intensity
away. As for the highest confidence value, namely at a 310lux and 959lux
70 cm camera height with a distance of 350 cm, 100 cm From the test data that has been obtained previously,
with a distance of 300 cm and a camera height of 140 an analysis of the fall detection system can be analyzed.
cm with a camera distance of 300 cm that is equal to In the fall detection system that has been designed and
100%. tested, camera position and light intensity greatly affect
the detection results of objects produced by the system.
The sensitivity equation indicated for the fall
detection system used in this study can be seen in
formula (3).

Figure 7: Graph of fall detection sensitivity value with


light intensity 959lux

4
Journal of Telecommunication, Electronic and Computer Engineering (JTEC)

intensity.
The confusion matrix equation that is indicated for the
fall detection system used in this study can be seen in
the formula accuracy (2), sensitivity (3), and error (4)
[11].

Tabel 1
Tabel Hasil Perhitungan

Figure 10: Comparison sensitivity graph in 310lux and


959lux tests The results of table 1, there are falling movements
detected fall by the system as many as 16 True Positive
In figure 10. note that the intensity of light is very data, falling movements but not detected by the system
influential in detecting falling objects. In the trial of amounted to 4 False Negative data, then non-falling
falling detection at a distance of 300 cm with a light movements detected by the system that fall detected
intensity of 959 lux has an average value of 100%, amounted to 3 False Positive data, and no falling
while at a distance of 300 cm camera with a light movements detected were not dropped by the system by
intensity of 310 lux only has an average value of falling 17 True Negative data.
detection of 80%. Tests presented in tabular form show real data. Where
In experiments using 310 lux light intensity, the in the fall detection system the results of testing with 40
change in distance between the camera and the object data show that the Accuracy of the system is 82.5%,
also has an effect on falling detection. This can be seen 80% sensitivity and 17.5% Error.
by the difficulty of detection by the system at testing
with a distance of 400 cm which is 40%. In addition, in
subsequent tests conducted there were 60% in VI. CONCLUSIONS AND SUGGESTION
experiments at a distance of 350 cm. So that the system
can only detect well at a distance of 300 cm which is A. Conclusion
80%. This can lead to the conclusion that the distance From the experimental results and the manufacture of
between the cameras systems in this study can be taken some conclusions as
with object testing affect the results of object follows:
detection carried out by the system. 1. The fall detection performance using Raspberry Pi 3
Then with a light intensity of 959 lux there is a high shows the maximum detection result at a distance of
average detection rate of 100% at a distance of 300 cm 300cm with a value of 100% and the state of the
camera. Due to the distance of the camera is not too far environment using a natural light source with a light
away on the object and is supported by environmental intensity of 959lux
conditions using natural light with the intensity of light 2. The fall detection performance using raspberry pi 3
received by the system of 959 lux. shows the maximum detection result at a distance of
From the data that has been obtained from testing with 300cm with a value of 80% as well as the state of
lux 310 or 959, changes in camera height made do not the environment using a natural light source with a
have an effect on in detecting falls, or it can be light intensity of 310lux
concluded that the height of the camera with 70 cm, 100 3. The intensity of light is very influential in detecting
cm and 140 cm has no effect on the value of confidence falling objects
object detection. 4. Tests with changes in camera height carried out at
70cm, 100cm and 140cm did not have an effect on
2. Analysis of falling and not falling data testing the detection of falls
Then a test is performed to calculate the accuracy, 5. In this dall detection study, it has an accuracy of
sensitivity and error in the fall detection system. Testing 82.5%, sensitivity 80%, and error 17.5% using trial
is done by taking a lot of data with different conditions, data falling and not falling.
namely falling and not falling. The data used are 100cm
camera height, 300cm camera distance and 959lux light

5
Journal of Telecommunication, Electronic and Computer Engineering (JTEC)

B. Suggestion [11] X. Kong and L. Meng, “Fall Detection for


In this research study, in order to become more Elderly Persons Using a Depth Camera,” vol.
leverage then given suggestions as follows: 13, pp. 269–273, 2017.
1. The system used still has a low accuracy of the
movement does not fall or fall movement. It is
recommended that further research improve the
system that has better accuracy than this research.
2. Necessary to improve the ability of hardware to be
able to detect falling people accurately and faster.
3. The hope is to be able to give an early warning if
the occupants of the house are falling, so that
residents can get help as soon as possible using an
email-based home care system application.

REFERENCES
[1] BPS, “Statistik Penduduk Lanjut Usia,” 2018.
[2] R. Istifada, “PADA PERAWAT DI
LAYANAN HOMECARE sangat luas ,
sehingga akses antara satu belum diaplikasikan
pada layanan,” vol. 5, no. 1, pp. 51–61, 2017.
[3] S. Norhabibah, W. A. K, and D. Risqiwati,
“Rancang Bangun Sistem Monitoring Deteksi
Jatuh untuk Manula dengan Menggunakan
Accelerometer,” JOINCS (Journal Informatics,
Network, Comput. Sci., vol. 1, no. 1, pp. 43–52,
2017.
[4] Rakhman, “Fall Detection System Using
Accelerometer and Gyroscope Based on
Smartphone,” pp. 99–104, 2014.
[5] S. Ishikawa, “Motion history image : Its
variants and applications,” no. June 2014, 2010.
[6] Mohd Brado frasetyo, “Comparison of Motion
History Image and Approximated Ellipse
Method in Human Fall Detection System,”
2019. [Online]. Available:
https://id.wikipedia.org/wiki/Ponsel_cerdas.
[7] A. F. Bobick and I. C. Society, “The
Recognition of Human Movement Using
Temporal Templates,” vol. 23, no. 3, pp. 257–
267, 2001.
[8] M. B. Frasetyo, J. T. Elektro, F. T. Industri, and
U. I. Indonesia, “Sistem pendeteksi jatuh pada
manusia berbasis pengolahan citra,” 2018.
[9] IPB, “CONFUSION MATRIX,” pp. 2–5, 2011.
[10] E. S. Wahyuni, M. B. Frasetyo, and H.
Setiawan, “Combination of motion history
image and approximated ellipse method for
human fall detection system,” Int. J. Simul.
Syst. Sci. Technol., vol. 19, no. 3, p. 13.1-13.6,
2018.

Das könnte Ihnen auch gefallen