Joint multi-target detection and tracking using conditional joint decision and estimation with OSPA-like cost

This paper addresses multitarget tracking (MTT) in clutter, including jointly detecting targets and estimating their states. Good solutions for MTT require solving the two problems jointly. A joint decision and estimation (JDE) framework based on a generalized Bayes risk was recently proposed for solving problems involving inter-dependent decision and estimation. In the JDE framework, a conditional JDE (CJDE) approach was proposed, which is conditioned on data. However, direct application of CJDE to MTT is difficult because the estimation cost for the case of multi-targets is not defined. The key to applying CJDE to MTT is to design a reasonable and tractable estimation cost. In this paper, we propose a CJDE risk that is inspired by the optimal subpattern assignment (OSPA), which is a widely used metric for MTT performance evaluation. OSPA unifies the estimation error of tracking and the cardinality error of detection, and has many nice properties. The proposed CJDE risk with the OSPA-like cost takes advantage of both OSPA and CJDE. Furthermore, this risk is not only reasonable but also easy to optimize. Based on this risk, we derive the optimal joint decision and estimation. For MTT, simulation results show that both the proposed CJDE and the existing recursive JDE (RJDE) outperform the traditional decision then estimation strategy in OSPA, and CJDE with the OSPA-like cost is better than RJDE in many cases.