Rapid computation of robust optical flow by efficient complementary voting

In this paper, we propose a novel rapid complementary voting-based optical flow estimation method and an original reliability criterion strategy for estimated optical flow. In order to estimate the optical flow of an interiest region, we divide it into some sub-references and then compute the similarity profile for each sub-reference by using a certain matching criterion. These similarity profiles can be used to exact two kinds of voting: positive voting ( candidate vectors) and negative voting (suppressing areas). The two kinds of voting are utilized to carry out complementary voting for obtaining an optimal estimation result of optical flow by adaptively contolling division of them. Moreover, according to the serious computation consumption in this algorithm, we also give out a sost reduction strategy called as PV-based Negative Voting. Experimental results with both sinthetic and real images demonstrate the effectiveness and praticality of the proposed algorithm.