Improving estimation of high accuracy optical flow by unstable region detection

This paper presents a new approach to estimate the optical flow problem. Many estimation methods have been proposed, but instabilities at boundaries of moving objects still presents a challenge in assessing optical flow. In this approach, we propose a new algorithm which is used to detect and readjust quickly unstable regions in result of estimating optical flow. Besides, we propose an anisotropic diffusion filter based on context. It is significant in processing intermediary data. Our approach is compared to other optical flow estimation methods taken over a number of real and synthetic sequences. Results indicate that the approach produces more accurate results for sequences with known ground truth.