Accurate dense optical flow estimation using adaptive structure tensors and a parametric model

An accurate optical flow estimation algorithm is proposed in this paper. By using a 3D structure tensor and a parametric flow model, the optical flow estimation is converted to generalized eigenvalue problem to avoid solving a linear system explicitly. The optical flow can be accurately estimated from the generalized eigenvectors. The confidence measurement derived from the generalized eigenvalues is used to adaptively adjust the coherent motion region to further improve the accuracy. Experiments on both synthetic sequences with ground truth and real sequences are used to test our method. Comparison with classical and recently published methods are given to demonstrate that our algorithm is accurate and robust to the aperture problem.