Moving objects extraction from video sequences based on GMM and watershed
Ren Ming-yi, Xiaofeng Li, LI Zai-ming · 2009
In this paper, a novel method for extracting moving objects from video sequences, which is based on Gaussian mixture model and watershed, is proposed. In order to overcome the drawback of subjective fixed threshold of traditional temporal segmentation, the difference image is modeled as a mixture of Gaussian distributions and a novel method to decide the model size and initial parameters of GMM is proposed. Then the expectation-maximization (EM) algorithm is fulfilled to obtain the Gaussian parameters and temporal moving area is detected; Considering the lack of traditional spatial segmentation algorithm of watershed, an improved watershed algorithm in accord with the human vision characteristics is proposed, it can restrain over-segmentation effectively; the temporal and spatial information fusion is fulfilled by ratio operation, and the video moving objects are obtained. Experimental results demonstrate the validity of the proposed algorithm.