Design of fall detection system using computer vision technique

Fall detection becomes an important topic in the homecare system. Compared to the wearable sensor, video-based fall detection system is more convenient and relaxed. In this paper, we propose a real-time and high accuracy fall detection system based on the video sensing stream. Firstly, we propose a fast and high-performance foreground segmentation method, which only uses the hue image to get the human information, and is it performed with simple adaptive background model. We also solve the problem of light and shadow change and achieve better results on PETS2001, PETS2006, and CDW 2014 dataset, compared with other algorithms. Based on this high performance segmentation technique, we can develop a fall detection system. The decision of fall detection is mainly based on the shape of human and the center of gravity. In our self-made falling sequences, the experiment results show that the accuracy of our proposed method can achieve 96% on average. Furthermore we can achieve real-time performance on embedded system with any GPU acceleration supported.