The research of extracting moving target based on GrabCut

Abstract

This study is designed to achieve a technology of extracting moving target within local windows based on GrabCut. After user cut out the moving target in the first frame, our system then covered a series of local windows over the contour of the target totally. Then our system used sparse optical flow algorithm and dense optical flow algorithm respectively to track the moving target and moved the local windows according the tracking result, which led to correct models and accurate extracting results. This study compared the affects to extracting results of tracking results between dense optical flow algorithm and sparse optical flow algorithm. Experimental results show that dense optical flow algorithm is better than the sparse optical flow algorithm and the former can ensure that the local windows cover the contour of target totally all along and lead to accurate extracting result of moving target.