Visual Tracking Based On Probabilistic Graphical Models

Visual object tracking is an important part of the current computer vision system. There are many visual object tracking algorithm such as optical flow method, kernel-based tracking, and bayesian tracking, etc. The existing tracking algorithms still have some difficulties in the face of the target obscured, non-rigid deformation and intense illumination changes and so on.For the past few years, with the development of probabilistic graphical model theory, people began to apply it in the visual object tracking. Different from the traditional visual object tracking algorithm, the probabilistic graphical model can express the spatial structure of the target in the process of modeling therefore provide a new research direction for the solution of the current visual object tracking problem.In this paper, it studies the visual object tracking algorithm is based on the theory of probabilistic graph model. Firstly, the approximate inference algorithms of the probabilistic graphical models were discussed, and then introduce particle filter tracking algorithm. Finally, we proposed two new visual object tracking algorithms. The first tracking algorithm based on hidden Markov models. By taking the characteristic description of the multi-block to calculate the weight of the particle, which result in the particle filter tracker to be more accurate. The second is based on the Markov random field theory. The algorithm track the multi-feature of the moving target through the independent particle filter between each frame, and then using the belief propagation algorithm to infer, determining the tracking results.