Research on Multi-target Tracking Algorithm Based on Classified Data Association

Data association, aiming at matching multiple targets between frames, is a key step in multi-target tracking. In this paper, we propose a multi-target tracking algorithm based on classified data association. We first use the RetinaNet detection algorithm to quickly obtain highly accurate detection results to provide reliable data for subsequent data association. Then, we build a model for the motion of the target and obtain the candidate association values of the target in the current frame according to the established model, thus eliminating the influence of interference detection values and reducing the possibility of false matches in data association. Finally, a classified data association method is proposed. The association scenarios are classified according to the association strength and number of target candidate associations, and specific data association methods are proposed for different association scenarios. The effectiveness of the proposed algorithm is confirmed by extensive experiments, and the experimental results show that the algorithm has high tracking accuracy and precision.