Content-aware optical flow estimation

While motion tracking is of use in many computer vision applications, optical flow estimation has become a challenging task that attracts vast interest in the past decade. In this paper, different from conventional pixel level energy formulation, we propose incorporating image structure analysis for content-aware optical flow estimation. In our content-aware method, image content analysis is carried out on the input frames, and spatiotemporal information is then extracted in the preprocessing procedure and incorporated in the proposed content-aware energy model to help enhance the accuracy of optical flow estimation. The experiment validated that the proposed content-aware method can attain a better accuracy in comparison with the content-blind model, and demonstrated better flow estimation results that were consistent with image structures.