A fast algorithm for moving objects detection based on model switching

A new method is proposed to improve background modeling speed. First, the pixels in current frame are classified into two classes according to average background to reduce the computing load. Second, different models for instance kernel or GMM based algorithm are used necessarily to deal with ´dead lock´ of scene. Third, a kernel density estimation based on neighbor correlation is used to decrease the false positives´. Last, the two algorithm detection results are fused to detect moving object by the label of pixel. In this paper, a novel description of correlation about the pixel with its around pixels and a strategy of background modeling are proposed. Experimental results of outdoor complex scene demonstrate that the new algorithm is robustness to noise and good for real-time moving object detection.