Research On Object Tracking Based On Graph-cuts And Detection

Moving target detection and tracking technology is a hot topic in the field of computer vision. At present, there still exist several problems in the process of target tracking:(1) dynamic changes of the background and its complexity;(2) the target occlusion;(3) the target deformation and its characteristics change. Aiming at the issues above, some improved methods are proposed in this paper, and the major work is as follows:First of all, we have analyzed and studied three main tracking categories: “detect-before-track” tracking, tracking based on dynamic segmentation and tracking based on distributions. On the basis of these analyses, in this paper, we address the problem of objects tracking by combining the advantages of the three classes of approaches. That is the target tracking method based on the detection and Graph- cuts technology is proposed.Secondly, this paper proposes an improved image segmentation algorithm based on Graph- cuts. The method adopts breadth-first search approach to search each pixel in the image, and the pixels belonging to the same area will be combined. In this way, we can overcome the time-consuming shortcomings when segmenting the larger pictures instead of using the traditional Graph-cuts algorithm.Then, in view of the different tracking situations such as object occlusion, deformation and etc, we should choose the appropriate detection algorithm to get the target observation. Then the predicted results and observations are added to the energy function, which will be minimized. Introducing the concept of observations can make target tracking results more accurate and able to handle the case of new target occurring. Experiments show that only using one energy function can't solve the problem of target fusion or occlusion. Aiming at this problem, other energy function is defined in the target merged areas. Finally the experimental results express that this proposed algorithm has strong robustness both for single objective and multiple target tracking.Finally, a target tracking algorithm based on Gaussian mixture model and Graph-cuts is presented. Aiming at different situations of the tracking target, this method will offer different treatment accordingly. A spatial-color Gaussian mixture model is built for each independent goal, and the method introduces into a strategy to replace the Gaussian mixture model for the occlusion target. Then energy function is built, and will be minimized by the Graph-cuts algorithm. Finally the experimental results show that the proposed algorithm has better performance and robustness to resolve the problem of the occlusion under the multiple objects tracking.