Variational optical flow estimation based on advanced data terms

Optical flow estimation is one of the key technologies in computer vision and image processing. However, constancy of the grey value which is used in traditional variational optical flow computation technology is sensitive to the constant changes of illumination and non-translational displacements. To solve this problem, the advanced data terms including the gradient value constancy assumptions and the laplacian constancy assumptions are introduced in this paper. And a flow-based smoothness term is introduced to preserve the edges of optical flow. Additionally, since the model strictly refrains from a linearization of these assumptions and coarse-to-fine approaches are capable to deal with large displacements. In the experiment, the efficiency and accuracy of improved algorithm is verified with some representative image sequences.