A method for real-time detection of human fall from video

In this paper we present a method for real-time detection of human fall from video for support of elderly people living alone in their homes. The detection algorithm has four steps: background estimation, extraction of moving objects, motion feature extraction, and fall detection. The detection is based on features that quantify dynamics of human motion and body orientation. The algorithms are implemented in C++ using the OpenCV library. The method is tested using a single camera and 20 test video recordings showing typical fall scenarios and regular household behaviour. The experimental results show 90% of human fall detection accuracy.