Multi‐Target Tracking Based on Two‐Layer Reinforcement Learning Optimization

ABSTRACT Data Association (DA) effectively matches sensor measurements with targets, which is crucial for the overall performance of Multi‐target Tracking (MTT). To address the DA problem in dense clutter environments for MTT, this paper proposes an MTT algorithm based on the two‐layer reinforcement learning (RL) optimization, which optimizes the DA while taking into account the influence of sensor positions on the DA process. In the presented algorithm, the first RL layer focuses on sensor path planning, enabling sensors to reach optimal observation positions at each time step to improve the measurement information of the target; the second RL layer optimizes DA decisions based on the refined measurements derived from previous optimal sensor path planning. This process involves training the weights of candidate measurements in the joint probability to better align with the motion characteristics of the targets, consequently improving filtering estimation performance. The simulation results show that the proposed algorithm efficiently enhances the estimation performance of MTT.