Fractional Variational Optical Flow Model for Motion Estimation

This paper introduces a new fractional variational optical flow model, which combines the fractional derivative with the variational optical flow method, for motion estimation. In the model, the data term is a weighted least-squares fit of local brightness constraint function and the prior term is a fractional derivative penalty function. So the proposed model can be seen as generalizations of integer-order optical flow model. We use the 2-D discrete Fourier transform to implement the numerical algorithm and propose an iterative scheme based on the coarse-to-fine warping strategy. Finally, we list various numerical results on six typical image sequences. Experiments demonstrate that the novel method is fairly insensitive to fractional-order variations and gives smaller angular errors for optical flow estimation.