Real Time Camera based Human Fall Detection Using Raspberry Pi

In today’s elderly population, fall is the major health complication problem. Every year,
around the world almost 30 percentage of falls occurs in elderly people aged more than 65.
When fall was occurred most of them need assistance to stand up promptly. The longer
they stays down on the floor after the fall the injuries becomes more severe as the time
passes, so they should be assisted as soon as possible. A fall detection algorithm based on
image processing techniques is so much useful in this situation.
This thesis makes use of OpenCV (Open Source Computer Vision) library for all the image
processing techniques in python language. This thesis develops a fall detection algorithm
and makes it possible in real time on Raspberry pi. In This system fall will be detected
based on parameters vertical velocity, aspect ratio, angle, traveled distance and change in
height of the elderly people and a mail will be sent through using SMTP protocol. Initially
the background will be obtained when the subject is not in frame and after that subject will
be tracked frame to frame, this tracking is useful in obtaining above parameters with
respect to last frame location. Fall will be confirmed when the change in above parameters
is abnormal. The partial objective human detection will be done using Histogram of
Oriented Gradients and Linear SVM method. Finally using SMTP protocol we will sends
an email whenever human fall is occurred. From the test results, this systems gives an
overall of precision, recall and total classification of accuracy is 93.75%, 88.23% and
90.65%.