Coarse-to-fine optical flow estimation with image structure tensor

This paper proposes a novel coarse-to-fine optical flow method based on the image local structure tensor. The data term of the optical flow model is projected by combining the proposed structure tensor constancy assumption and the grey value constancy assumption with the robust penalty function, and an isotropic nonlinear function is planned to smoothing term. With the coarse-to-fine multiscale warping strategy, a linear iteration scheme is established. Several experimental results of the Middlebury sequences show that the presented method is high accuracy and good robustness.