Post-Fall Time Accounting for Fall Detection Using a Portable Camera

Abstract

Falls are one the major problems that threatens the health of the elderly. For this reason, many devices have been developed by researchers all around the globe to continuously monitor and detect critical events, like falls, which allow for a fast-medical intervention to take place. The proposed method for the detection of fall is based on the original version of the Histogram of Oriented Gradient (HOG) combined with Optical Flow and immobilization time to reduce the numbers of false detections. The method was implemented in a system composed of a portable camera and an embedded multi-core computer (Raspberry Pi) to parallelize computations which allows for real time detection. The results of 45 tests conducted on 09 subjects show that falls from standing position can be detected with 80% of sensitivity. The inclusion of immobilization time in the detection process improves the specificity for rotations by 14%.