Simultaneous optical flow and occlusion estimation for images with large displacements

Abstract

In this paper we present a method for estimating occlusion of pixels in image sequences. The estimation of the occlusion layer is based on optical flow. In our proposal the optical flow is computed in an exhaustive way by comparing neighborhoods between consecutive images. In order to compute the optical flow we consider three consecutive images (previous, actual and next frame). The optical flow is computed backward (between actual and previous frame) and forward (between actual and next frame). The proposed method considers an energy model which is minimized. The Euler-Lagrange equations let us to compute backward optical flow, forward optical flow and the occlusion layer. The comparison between neighborhoods is stored in a cost volume avoiding computing neighbor comparison repeatedly.