Fast motion region segmentation based on motion vector field

Abstract

In this paper, a fast method for extracting motion object regions from moving background is proposed. The proposed approach estimates global motion parameters and implements global motion compensation (GMC) using the motion vector (MV) field. According to the predictive errors of MVs generated by GMC and the movement difference between the background and foreground regions, a dual- threshold mechanism is employed to perform initial region classification. Computing the centroids of the background and foreground set, a recursive cluster procedure is applied to segment the motion object regions. Experimental results show that the proposed method can segment moving regions not only fast but also fairly accurately.