Optical flow estimation with parameterized data term and warping

Horn/Schunck approach which has been widely used in variational optical flow estimation consists of a data term and a smoothness term. In this paper, a parameterized data term is proposed. The parameter can adjust the proportion of the two images´ gradient for obtaining the Euler-Lagrange equations. This combination makes the optical flow more robust to noise and illumination changes. Firstly, the classic model is analyzed, especially the coefficients of the Euler-Lagrange equations. Then, a model with parameterized data term is proposed. This model has formulated the combination of image gradients theoretically and enabled more flexible combination. Finally, a multi-resolution technology has been used for solving the nonlinearity of optical flow. In this process, the warping step is also parameterized and the influence of the parameter is analyzed. The experiments demonstrate the benefit of our parameterized model to the classical one.