TY  - JOUR
T1  - Improved Elderly Fall Detection by Surveillance Video using Real-Time
Human Motion Analysis
AU - Dorgham, O. AU - Rass, Sanad Abu AU - Alkhraisat, Habes 
JO  - International Journal of Soft Computing
VL  - 12
IS  - 4
SP  - 253
EP  - 262
PY  - 2017
DA  - 2001/08/19
SN  - 1816-9503
DO  - ijscomp.2017.253.262
UR  - https://makhillpublications.co/view-article.php?doi=ijscomp.2017.253.262
KW  - Fall detection
KW  -real-time analysis
KW  -human activity recognition
KW  -surveillance video
KW  -object classification
AB  - Statistical studies show that around 28-35% of older people aged 65 and over fall each year. This
percentage increases to 32-42% among those over 70 years of age. These figures explain the dramatic increase
in the number of systems that have been developed in recent years with aim of detecting falls. In this study,
we propose, implement and evaluate a multiphase system framework to analyze human motion in real time to
detect falls among the elderly. The system phases consist of background subtraction to extract the foreground
of the frame for further analysis object classification which performs some morphological operations and draws
the contours of the detected objects to identify human bodies object tracking to reduce false alarms and fall
detection to detect the occurrence of falls based on the bounding rectangle and surrounding points contour
drawing methods and by utilizing dual-camera verification as well as a method of leg detection using a camera
situated above the subject. The system starts with a background subtraction phase to detect moving objects.
After that the moving object is classified as a human body or not. Objects classified as human bodies are then
tracked to detect falls. The experimental results showed that the system has high performance and accuracy
and that it can implement and process live videos and report falls instantly. Our system can process videos at
an approximate frame rate of 20 FPS (using an Intel 2.8 GH quad core processor with 4 GB RAM) and with an
accuracy of 88.1%.
ER  -