Gaussian mixture model for background based automatic fall detection
Huer Xiao, Zhiliang Wang, Qiang Li, Xianmei Wang · 2013
It's very dangerous for the elderly to fall, so fall detection is very important in nowadays society. This paper addresses to detect fall activities by combing Gaussian mixture model (GMM) and special-temporal analysis of aspect ratio. First, we use GMM to get the background part and foreground part from an image. After morphological operations, some small gaps are removed by empirical knowledge from the foreground part. Second, we calculate the aspect ratio feature from the minimum external rectangle of a human body. Through the spatial-temporal analysis of aspect ratio, we output the fall behaviour more robust. The experiments show that our approach can effectively detect human falls in real time.