Performance Evaluation of Bayesian Multi-target Trackers

Bayesian multi-target tracking algorithms have attracted substantial interest in recent years for their theoreticalconsistency. While much attention has been devoted to the design of Bayesian multi-target trackers, little work has been presented for their performance evaluation. Since no explicit association between tracks and targets are made, the performance evaluation of Bayesian multi-target tracking algorithms is difficult. This paper presents a scheme for this purpose. An optimal assignment algorithm to associate the estimates with the truth is proposed. Based on the estimate-to-truth assignment, we adopt a robust method for the track-to-truth assignment. Finally, several measures of effectiveness (MoEs) on track level can be defined to take a closer look at different aspects of Bayesian multi-target tracking algorithms. Simulation results show the effectiveness of this evaluation scheme.