Performance of PHD and CPHD filtering versus JIPDA for bearings-only multi-target tracking

The performance of three multi-target tracking algorithms are compared under the challenging problem of bearings-only tracking in the presence of clutter and missed detections. The algorithms under consideration are the Gaussian Mixture Probability Hypothesis Density (GMPHD) filter, the Gaussian Mixture Cardinalised Probability Hypothesis Density (GMCPHD) filter and the Joint Integrated Probabilistic Data Association (JIPDA) filter. A Monte Carlo analysis is presented for a difficult bearings-only tracking scenario, in which the algorithms assume a diffuse model for target birth, such that new targets may appear at any bearing and at any time. The algorithms are evaluated in terms of the Optimal Sub-Pattern Assignment (OSPA) metric, the cardinality estimation performance, and their respective computational requirements.