Simultaneous occlusion handling and optical flow estimation

Abstract

Occlusion handling and optical flow estimation is a chicken-and-egg problem. In this paper, we propose our method which can handle occlusion and estimate optical flow simultaneously. First of all, we use the backward interpolation strategy to gain the warped image, then we obtain the occlusion relationship by comparing the pixels before the movement. After that we use the occlusion relation to correct the warped image and get the occlusion coefficient. Later, using the occlusion coefficient to modify the energy function. Finally, the corrected energy function and warped image are used to estimate the final optical flow results. We evaluate our method on some popular datasets such as Flying Chairs and MPI-Sintel. Experimental results demonstrate that the proposed method improves the accuracy of current optical flow estimation methods significantly.