A Track Quality Based Metric for Evaluating Performance of Multitarget Filters

The track qualities of multitarget state estimates are often available in many tracking algorithms, including the multiple hypothesis tracking and the joint integrated probabilistic data association estimators. However, the recently proposed Optimal Subpattern Assignment (OSPA) metric ignores the quality information and thus its capability to quantify the performance of multi-object estimation algorithms is limited. In this paper, a new metric, called the quality-based OSPA (Q-OSPA), is proposed based on the original OSPA metric. The proposed Q-OSPA metric is able to incorporate the quality information and thus provide more accurate quantification of the performance of multi-object estimation algorithms. Also, the mathematical consistency of the original OSPA metric is maintained by the proposed Q-OSPA metric. In addition, if the qualities of estimates are not available, the Q-OSPA metric reduces to the original OSPA metric by assigning equal qualities to the estimates. Besides theoretical derivations, simulations are presented to verify the advantages of the proposed metric.