Automatic segmentation of moving objects in video sequences based on spatio-temporal information

An image segmentation method for separating moving objects from background in video sequences is addressed in this paper. The proposed method utilizes spatiotemporal information. Firstly, the global motion is estimated, in which adaptive rood pattern search (ARPS) method is adopted to search the best match block, decreasing computation load. Then, colour information is incorporated into getting the frame difference. For getting the initial contour of moving objects in the image sequence, two consecutive image frames are examined and a hypothesis testing is preformed by comparing two variance estimates from the frame difference of two consecutive images, which results in an F-test, indicating moving areas(foreground) and nonmoving areas(background). Lastly, an improved active contour algorithm, gradient vector flow (GVF) snake, is performed to refine the initial contour and to find precise object boundary eventually. This paper presents various experimental results. Simulation results show that the proposed method gives better performance in terms of the computational efficiency and the segmentation accuracy.