A variational Bayesian approach for partly resolvable group tracking

Abstract

The partly resolvable group tracking is significant for the anti multi-agent system, in which some individual targets of the group may generate only one measurement, and the association relationships between the individual targets and measurements are unknown. To estimate the group state, the individual target states, and the association variables accurately, this paper proposed a novel tracking algorithm considering the resolvability of group. Firstly, a unified Bayesian framework for partly resolvable group tracking is formulated, including data association and state estimation. Secondly, a variational Bayesian approach is derived using the Kalman filter and random matrix model, which recursively estimates the states of the group and its individual targets. Thirdly, to reduce the computational burden, the complex data association problem is modeled by a factor graph that is solved by message passing. Finally, simulation results show that the proposed algorithm has a significant improvement over the previous variational group tracking algorithm in terms of state estimation.