Multi-Sensor Management for Multi-Target Tracking Using Mutual Information.

When multi-sensor systems are used for tracking multiple targets, adaptive sensor management is critical to ensure high precision and efficiency of data processing. This paper presents a novel sensor management method for automatically selecting informative sensors from the multi-sensor system so as to improve tracking performance. Our proposed solution, uses a Partially Observed Markov Decision Process (POMDP) formulation with the labeled multi-Bernoulli filter for tracking. The sensor management problem is posed as a as a constrained POMDP with an information-driven objective function. For numerical tractability, the heuristic greedy algorithm is used to search for the suboptimal solution to the POMDP. Simulation results validate that the proposed method provides satisfactory tracking accuracy while reduces the number of selected sensors.