Motion detection based on spatio-temporal saliency perception

This paper proposes a method of motion detection which fuses spatial and temporal saliency information. First, by introducing the edge information to the Itti model, after extracting the brightness, color, direction and edge features, the spatial saliency map is obtained by fusing these information with center-surround mechanism and Gaussian difference. Second, motion saliency in temporal domain is extracted by applying kernel density estimation to video sequence. Finally according to the motion preference rule and computing the two saliency information weights, the spatiotemporal salient map can be computed and the moving target can be detected. Experimental results show that the method with the fusion of spatio-temporal saliency can detect the moving object exactly and effectively.