A Stochastic Approach Based Bayesian Salient Motion Segmentation Method

Moving object segmentation techniques are fundamental and crucial for video surveillance. In this paper, we abandoned the traditional background differential model approach which ignores the foreground modeling, and address this problem under the Bayesian framework. Our major contribution can be summarized as, modeling the background and foreground competitively to augment the segmentation accuracy, implementing the prior information to enforce spatial-temporal consistency, and introducing the Monte Carlo importance sampling techniques which effectively reduces the computation complexity while guarantees the expected veracity. Promising results demonstrate the potentials of the proposed framework.