Abstract: In this paper, we propose a new method to estimate the optical flow fields of a sequence of images. With the assumption that the velocity is constant from frame to frame along with the flow vectors, which is very common in real scene, we develop a new constraints on the flow field. We formalize the estimation of flow fields as an energy minimization problem and adapt the iterative reweighted least square (IRLS) to solve it. To make the penalties robust to outliers, we mainly consider the Charbonnier penalty. We follow the modern optimization procedure including coarse-to-fine schema, graduated non-convex (GNC) schema to get the final result. The effectiveness of our method is borne out by experiments on both synthetic and real scene.