This article provides a broader validation of the Mahalanobis distance-based M-OSPA method for evaluating the performance of multiple extended object tracking algorithms. Modern embedded sensors for environment perception have significantly improved resolution. Accordingly, they permit an extended representation of the targets of interest, where multiple points are distinguished and separately tracked per target during its motion in the sensor’s field of view. The M-OSPA metric is applied for several random finite set (RFS) based tracking algorithms, in order to demonstrate its capacity to select the optimal tracking algorithm. The simulation results show that it can be used as a powerful instrument to identify the most performant multiple object tracker in state-of-the-art automotive applications.

Keywords: OSPA metric, PHD filter, Multi-Bernoulli filter, Radar