Optical Flow Estimation Method Based On The Variational Theory

Optical flow estimation is an important research topic in computer vision. In recent years, optical flow has attracted more domestic and international scholars and received broad attention in the field of computer vision.This dissertation focuses on optical flow estimation method based on the theory of variation. To deal with the fact that the noise of the illumination change and optical flow solution efficiency in the traditional HS algorithm, an improved optical flow estimation based on image decomposition and multiple grid method is presented. For the shortcomings of the over-smoothing in L2 norm of the traditional HS model, L1 norm is introduced. TV-L1 variational model based optical flow field estimation method is adopted to improve the precision. To further improve the accuracy of the optical flow model and solve the problem of large displacement optical flow, on the basis of TV-L1 model, variational optical flow field estimation method based on CLG is adopted in this paper. The main work is as follows:1. An improved optical flow estimation based on image decomposition and multiple grid method.The estimation of optical flow suffers from its sensitivity nature to illumination change, its high computational complexity and slow convergence property, which severely compromise its performance. In order to tackle the aforementioned problems, an image decomposition technique based on ROF model is introduced to deal with varying illuminations, and a multi-grid based optical flow hierarchy strategy is presented for a fast implementation. We propose to distribute the iterative procedure on several grid layers with different resolutions to obtain a fast convergence and in turn speed up the optical flow computation. The finer grid promises to eliminate higher frequency errors while the coarser level is employed to cope with lower frequency components.2. The optical flow field estimation method based on TV-L1 variational model.For the problem of the sensitivity to noise and the over-smoothing in L2 norm of the traditional model, a first-order items(L1 norm) is introduced in optical flow model, which can well-maintain the piecewise smooth and improve the accuracy of optical flow calculation. During linearization process, the process of obtaining the optical flow field is limited, for the first-order Taylor expansion can be only applied to small displacement projected coordinate. To solve this problem, coarse-to-fine pyramid algorithm is introduced. Using multiple scales method, iterative linearization method can not only be directly employed, but also avoid iteration convergence to local minimum.3. Improved method for variational optical flow field estimation based on CLG.To suppress the impact on optical flow estimation caused by noise and solve the large displacement optical flow, a CLG-TV variational optical flow estimation model based on CLG(Combined Local and Global) method is established. CLG method is utilized to integrate TV-L1 model with LK algorithm, which is the local optical flow estimation method, and anisotropic diffusion and bilateral filtering technology are adopted in CLG-TV model. On the basis of the TV-L1 model advantages, this model has high robustness to noise and ability to solve large displacement problem. And structure and texture decomposition method is integrated with coarse-to-fine approach during the solution process to improve the accuracy of optical flow computation.Experimental results show that the improved optical flow estimation based on image decomposition and multiple grid method, the optical flow field estimation based on TV-L1 variational model, and the improved method for variational optical flow field estimation based on CLG, which are presented in this article, are effective for optical flow estimation. They have theoretical significance and practical application for researches on motion image analysis and target tracking.