Research On Target Tracking Algorithm Based On Image Sequences

Object tracking based on image sequences is a key research direction in computer vision field, which involves many subjects. In numerous algorithms, the Mean Shift algorithm is a very good algorithm in object tracking field, which uses kernel function to match the object model and candidate object model so that the algorithm has a certain robustness to the change of the moving state of the object, and it is also fast, efficient and easy to be integrated with other algorithms. But, the algorithm only uses the color distribution characteristics of the tracking window to establish the model, when the tracking environment is more complex, such as the speed of the object is fast and the case of occlusion, the tracking effect is poor.Therefore, it is very meaningful to study on the Mean Shift algorithm and improve the tracking effect.This paper firstly introduces the research background and significance of target tracking algorithm based on image sequences and the research status, analyzes some shortcomings of the traditional Mean Shift algorithm and proposes the improved algorithm. This paper mainly studies on the following several aspects:1)The traditional algorithm has fixed bandwidth so it has poor tracking effect of variable size target. A bandwidth adaptive algorithm based on affine transformation is proposed. Firstly match the target center of two consecutive frames by backward tracking. Then the target is matched with the method of corner matching to adjust the size of the tracking window. The experiment shows that the proposed algorithm can achieve accurate tracking in real time when the target size is changing.2)The tracking effect is not good when the speed of the target is fast or a serious occlusion occurs. Based on the bandwidth adaptive algorithm, this paper proposed an algorithm that combines the Mean Shift algorithm with the kalman filter. Before the iteration in every frame, the kalman filter predicts the possible position of the target in current frame by the motion state information of the target in last frame, and take this position as the starting position of the iteration. Because the algorithm has a certain predictive function, it solved the problem of the target information lost of the tracking window. The experiments show that the improved algorithm is effective for solving the above problems.