Efficient Computation of Optical Flow Using Complementary Voting

This paper describes an efficient complementary voting based estimation algorithm of optical flow with reliability measure, called CMV. To estimate the optical flow of an interest region, we divide the region into subreferences, then compute the similarity profile for each subreference using a certain matching criterion. These similarity profiles are employed to extract two kinds of voting roles — positive voting and negative voting. Subsequently, the two kinds of voting roles are carried out to obtain an optimal voting result which is used to estimate an optical flow and a reliability value under a control criterion. To reduce the computational complexity of the CMV, we propose a PV-based negative voting strategy. Experimental results show that the CMV is effective for estimating optical flow in poor-quality image sequence, and that the new negative voting strategy greatly reduces the computation complexity without degradation of the performance.