A novel spatial-temporal optical flow method for estimating the velocity fields of a fluid sequence

摘要

In this paper, we propose a novel optical flow method to estimate the velocity fields of fluid sequences based on spatial-temporal physical principles. Our novel optical flow model mainly takes scalar field correspondence, velocity field temporal coherence and incompressibility into consideration. And the velocity field smoothness assumption is also used for better convergence. Compared with existing methods which only estimate the velocity field between a single pair of frames, our novel optical flow model can deal with the fields of all frames of the sequence simultaneously. For some frames of the sequence, the image quality is higher and more accurate velocity fields can be obtained, but for other frames, the estimated fields may be rather far from the groundtruth due to the lack of information. And our model can propagate the correspondence information from the accurate frames to neighbor frames and help the optimization converge to a better result. Also, several sophisticated optimization techniques are employed to solve our novel model efficiently and robustly.