Abstract
Most proposed methods of detecting moving objects can not segment the players in soccer videos well. In this paper, we present a novel background subtraction algorithm based on Gaussian Mixture Models (GMMs). In the initial stage, we pre-process the frame with compensation construction method we propose. In the updating stage, the moving area and background one can be classified by differencing two consecutive frames. According to the special area, we update the GMMs with corresponding rate. Experimental results show that our algorithms are effective in detecting moving objects in the dynamic scene.