An efficient algorithm for estimating the inverse optical flow

Abstract

The aim of this work is to propose a method for computing the inverse optical flow between two frames in a sequence. We assume that the image registration has already been obtained in one direction and we need to estimate the mapping in the opposite direction. The direct and inverse mappings can easily be related through a simple warping formula, which allows us to propose a fast and efficient algorithm. Nevertheless, it is not possible to estimate the inverse function in the whole domain due to the presence of occlusions and disocclusions. Occlusions occur because some objects move to the same position in the next frame. On the other hand, disocclusions are the opposite situation: no correspondences can be established because there is no object moving to those positions. In this case, there is a lack of information and the best we can do is to guess their values from the surrounding values. In the experimental results, we use standard synthetic sequences to study the performance of the proposed method, and show that it yields very accurate solutions.