A Study Of Optical Flow Computation Technology Based On Variational Theory

Optical flow shows the speed of the pixel movement in the observation surface of spacial mobiles. In fact optical flow does not only contain movement information of objects observed, but also carries abundant information of 3D structure features. The optical flow computation is a key technique in the fields of Computer Vision and Image Processing, and become the research hotspot today, so that this paper will study and improve the optical flow computation technology based on variational theory.The start of variational optical flow computation technology is to transform the acquisition of optical flow field to the minimization of a global energy function. First of all this paper demonstrates the transformation of the minimization of energy functions into the nonlinear partial differential equation in mathematical way, and analyzed the consistency of solving the diffusion-reaction equation with minimization of the energy functions.Then this paper improves the data term and the smoothness term separately , and propose a new diffusion-reaction equation. For data term, this paper combines two kinds of constancy assumptions. Based on the brightness constancy assumption,the improved data term introduces the Hessian constancy assumption with more image information, enhances the accuracy of optical flow computation when illumination varies unexpectedly, and meanwhile elevates algorithmic robustness in noise by combining with the Lucas' local constancy assumption. For smoothness term, this paper designed the diffusion term in diffusion-reaction equation directly, so that it is unnecessary to consider the restrictions on convexity and differentiability, and can have more flexible control on movement edge. On the process of smoothing optical flow , this paper does not only applies Weickert's coherent enhancing diffusion theory, but also extends it to the optical flow vector field and make some amendments more suitable for optical flow field. In order to solve the effect of large displacement errors on optical flow computation, this paper makes computation in the coarse-to-fine warping strategy.Finally, this paper selects three representative image sequences to test effectiveness of the improved algorithm. It's proved that the algorithm proposed in this paper is superior to the previous variational algorithm in robustness and accuracy, by analyzing the optical flow vectors and error data which are both obtained by experiments. .