Automatic moving object segmentation in video sequence

Abstract

A novel algorithm for automatic segmentation of moving objects from video sequences is proposed in this paper. Firstly, a fourth-orders statistic hypothesis testing in inter-frame difference is used to automatically separate the moving areas from the background in a general video sequence. Then the morphological filter is used to remove the noises and holes. Secondly, for three consecutive frames, superposition of the moving areas of first and second frame and that of second and third frame highlights the moving object areas of the second frame, the covered/uncovered background may be eliminated using such idea. Thirdly, the initial contour of the moving objects is determined by extracting the contour of the moving object areas and an improved active contour which uses the gradient vector as the external force guides the initial contour moving to the accurate moving object contour. Lastly, the motion compensation which uses the new three-step search algorithm is made to the moving object contour to obtain the initial contour of the next frame and the improved active contour is used again to correct the moving object contour of the next frame. Thus, following frames segmentation can be realized. Experiment testifies that the proposed algorithm is of few parameters, robust to noise and best in result of segmentation.