Selective block search for surveillance centric motion estimation

In this paper, we propose a novel approach to perform the selective motion estimation specific to surveillance videos. A real-time background subtractor is used in which pixels representing motion activity are bounded in multiple rectangular boxes. Dimensions and coordinates of each bounding box are used to locate these pixels in the motion estimation module. Motion estimation is performed only for those blocks which overlap with bounding box. This improves utilization efficiency of computing resources by focusing on pixels which are important from surveillance standpoint. Experimental evaluation shows that significant reduction in computational complexity can be achieved by applying the proposed approach.