Research On Moving Object Detection Algorithm Based On Variational Optical Flow Field

Moving target detection is an important subject in the study area of computer vision, which has broad prospects for application in many fields of intelligent transportation, security forces and so on. Because the optical flow not only contains the movement information of objects to be observed, but also carries the important information to observe the three-dimensional structure of objects, this technique has become one of the important methods of the moving target detection. Practical applied results show that the optical flow method to processing the moving target detection is easily affected by the change of illumination, and has some other disadvantages of low noise immunity, poor stability, large amount of computation, inaccurate calculation of moving edge, when handles moving target detection, how to effectively improve the optical flow method for reliability, stability, real-time of application to detect the moving target has become one of the hot topics in computer vision research field.In accordance with the lack of optical flow method of application to detect the moving target, this paper propose improved variational optical flow method, the results of the computer simulation experiment show the feasibility and validity of the method to propose.The main work of this paper is as follows.(1)We have studied the optical flow computation technology based on variational theory. we changed the energy functional into a nonlinear partial differential equation and detected it with mathematical method optical, then analyzed the consistency of diffusion reaction equations and minimization energy functional. improved the data items and smoothing items in the energy functional, obtained the new diffusion reaction equation.(2)The data items are improved by the combination of the Gray conservation hypothesis and the Laplacian conservation assumption, and effectively increase the accuracy of optical flow computation for light mutation case. the combination of the Lucas local constraint algorithm is effective to increase the robustness of the algorithm in the presence of noise. By introducing the smoothing term based on isotropic andimage driven to achieve the purpose of eliminating noise and protecting edge. this paper solves the influence of large displacement problem on optical flow computation by means of multi resolution and multi resolution method.(3)Proposed a new moving target detection method which combined the improved optical flow method with three frame difference method. After calculating the optical flow of the Harris corner points by using the improved optical flow method which is the important and representative local feature points, introduced the binarization three frame difference images as a supplement to simplify the optical flow method.Experimental results show that the proposed method has obvious advantages compare with the two methods in terms of improving real-time requirements to compute optical flow.(4)Considering the disadvantages of binary-value image with fixed threshold. By comparing and analysing the methods of common adaptive threshold selection, select the adaptive threshold with OTSU as binarization threshold that Binary-conversion for three frame difference method, by searching for threshold that meet maximum-between-clusters-variance,it makes some pixels of foreground image which three frame difference method receive generate more accurate result, Experimental results show that the range of detection moving targets is more complete with this method, because OTSU has quick, easy and stable features, so it can be raised greatly in the field of computational efficiency.