Optical Flow Calculation Methods, Duality

Optic flow can describe the gray-scale trends of each point in an image sequence. Its reliable computation has become one of the main challenges in computer vision, and variational methods belong to the most successful techniques for optic computation. One of the research content of this article is that dual method is presented for small optic flow computation based on variational models. Data terms of variational models can base on the assumption of different light conditions and different motion types, such as base on constancy assumptions of the gradient, the Hessian, the gradient magnitude, the Laplacian and the Hessian determinant. Robust estimation form of L1 norm is used in data terms and the regularized terms use total variation of optic flows with edge preserving property. The dual variables for total variation terms avoid using small regularizing parameters in classic total variation terms and improve accuracy. Experiments show the effectiveness of the method. This method can be extended to computation of large displacement optical flow and higher order models. Optic flow computation is divided into small displacement and large displacement optical estimation. Small optic flow refers to the displacement of an object is small in motion(less than or equal to one pixel), but the movement of objects in reality is often very quick, and the displacement of adjacent objects is very large. So the research of large displacement optical estimation has important practical significance. Another research content of this article is that we use dual method for large optic flow computation based on the study of small optic flow computation, and discuss fully the numeric arithmetic. Also the impact is illustrated by experiments finally.