| As a novel and rising subject in computer vision, Video tracking includes many areas such as signal and video processing, communication and computer vision. It is defined as detection, extraction, recognition and tracking of the moving object in video sequences in order to get the parameters of the movement such as location and velocity, which are used for further processing and analysis. The key of object tracking lies on the correspondence between the tracking object and the static object.There is a popular problem that object tracking will be missing or inaccurate, in the case of occlusion and large appearance variation. Aim at tackling this problem, we proposed a fragment-Based multi-feature object tracking method. Our method applied the particle swarm optimization (PSO) algorithm and made the most of its high efficiency and less parameters advantages. Color histogram and histogram of oriental gradient are applied to represent the object, and we employ Bhattacharyya distance and chi-square distance to measure the similarity of the object; through PSO algorithm, we attain the initial position of object, and then combine Kalman filter and adaptive scale adjustment to locate the accurate position. In addition, we make full use of target block division to deal with the occlusion problem effectively. We conduct extensive experiments on standard video and our own video. The experimental results shows that our proposed approach is robust for tracking moving object.In addition, this paper systematically introduce and conclude the particle swarm optimization algorithm. Though a mass of extensive and intensive technical reading, we did a careful study on its research background, relative studies and future work. We also compared different improved PSO algorithms, and conclude their advantages and disadvantages. Based on the comparison and considering our demand, we adopt a inertial-weighted PSO algorithm, which is validated to be very efficient on object tracking in our experiments. |
PDF Full Text Request