Improving the Optical Flow Accuracy Based on the Total Variation of Local-Global method

Yugui Zhang, Haonan Fan, Jin Zheng, Chi Zhang

Abstract: The aim of this paper is to improve the optical flow accuracy for the preservation of flow discontinuities in total variation method. The proposed method is based on the total variation of the local-global method, including the following steps: firstly, the initial optical flow is obtained from the local-global method; secondly, the data item and the smoothing item are designed based on initial optical flow by using the total variation, with the data item involving the brightness constancy and the gradient constancy, and the smoothing item involving exponential function; and finally, the optical flow results from the second step are optimized by using median filtering. The experimental results show that our proposed method could enhance the accuracy and robustness of the optical flow on the Middlebury Dataset and the MPI Sintel Dataset.