Research on Multi object Occlusion Tracking and Trajectory Prediction Method Based on Deep Reinforcement Learning

In order to improve the accuracy of multi-target occlusion tracking and trajectory prediction in complex scenes, an ensemble method based on deep reinforcement learning is designed. By constructing three core modules of occlusion detection, deep reinforcement learning tracking, and trajectory prediction, and utilizing spatiotemporal feature fusion and attention mechanisms to achieve multi module collaboration. The experiment was conducted on datasets such as MOT17 and MOT20, and the results showed that the proposed method had significantly better target recovery ability and trajectory prediction accuracy in occluded states than traditional methods. The accuracy of multi-target tracking was improved by 12.3%, and the overlap of long-term trajectory prediction was increased to 0.72, verifying the robustness and effectiveness of the proposed method in complex dynamic environments.