Human body fall detection based on the Kinect sensor

Health problems of the elderly are more and more serious with the growth of aging population. The accidents like falling down need to be paid more attention especially. Through the analysis of the existing detection methods, a more simple and rapid algorithm about human body fall detection based on the Kinect sensor is proposed in this paper. This algorithm is composed of three parts, which are moving target depth of image acquisition, processing of depth image and identification of target motion behavior. The realization of the detection algorithm is based on the depth map sequence obtained by the Kinect. The data of falling and bending are collected and compared in this experiment. And the OTSU algorithm which has anti-noise performance is used to process depth map. It is conducive to the body contour extraction. After extracting the contour, the corrosion operation is used to repair the edge. Then three parameters are extracted, which are the aspect ratio of human external rectangle, the gravity center of human body and the inclination degree. Every frame image outputs aspect ratio and dip angle value. By comparing these values with the threshold, the system judges whether human falls down. The experimental results show that this algorithm is a kind of effective fall detection algorithm.