Boundary extraction of moving object based on spatio-temporal information

This paper propose a novel algorithm of boundary extraction of moving objects. This method utilizes spatio-temporal information. Firstly, a hypothesis test is performed to detect the change area, that is, comparing two variance estimates, respectively obtained in a observation window and in background from frame difference of two consecutive images, results in a X-2-test. After some necessary morphological, scanning and filling-in operations, an area including foreground and little background is gotten. Then Canny operator is adopted to get its edge image E, consisting of the moving object's edges and very few edges from the background. For getting the boundary of the target, a sequence of morphological technology is used. A connecting method makes sure the object's boundary is intact and connected, in which, a 3*3 mask is utilized to connect pixels which possibly belong to the boundary based on the direction information. Then a disconnection method and a pruning method are adopted to remove the parasitic components and a threshold value is set for connected area to remove the edges due to textures inside the object or in background. Finally, we can get the exact and smooth boundary of moving objects. Simulation results show that the given algorithm performs well in terms of the computational efficiency and the segmentation accuracy.