Video Multi-Target Tracking Based On Association And Filter

Video Multi-target tracking is one of the core issues in the field of computer vision, and merges advanced technologies in many subjects such as image processing, automatic control, pattern recognition and artificial intelligence. It is widely used in military, video surveillance, traffic control and human-computer interaction, and other fields, has wide application prospect and great economic value.Based on the static, single camera video sequence, the paper had extensive research and simulation analysis on the video object detection, filtering algorithms and multi-target association in video multi-target tracking. First of all, frame difference and background subtraction base on the average were introduced, and Gaussian mixture model (GMM) is studied, and shadow detection based on different threshold of two color space was used and got a good test effect. Secondly, kalman filter algorithm which adapts the linear Gaussian was introduced, and focused on the study of particle filter for nonlinear non-Gaussian system, which is the theory basis of subsequent video tracking. Thirdly, the association algorithm in multi-target tracking was introduced, and for the features of video, the method, based on the combination of the Euclidean distance and color histogram distance, which can give a reasonable match to the trajectory and determine the occurrence of conflict situation, was selected. Finally, on the basis of front knowledge, an adaptive filter tracking algorithm base on particle filter was proposed. The algorithm determines the conflict situation based on multi-target association first, it chooses the observation model based on color with no conflict, and chooses the observation model based on the association of color and spatial information with conflict. The algorithm not only allocates compute cost reasonable, but also effectively handles the complex motion between multiple targets such as occlusion, splitting and mergence.