Free-Clustering Optimal Particle PHD Filter for Multi-target Tracking

Analytical expression of the optimal sampling density for the particle probability hypothesis density (PHD) filter is difficult to obtain, and the filter performance degradation is caused by clustering to extract the targets state. A free-clustering optimal particle PHD filter algorithm is proposed in this paper. This study found that, the particles at previous time step in general has a great relationship with certain measurement in the next measurement set, while little relationship with the rest measurements, so the optimal sampling density can approximate to the form related to a single measurement. After linearization of the measurement equation, the approximate analytical form of the optimal sampling density can be obtained. For the association between particles and measurements, the particle has class information, so the clustering algorithm to extract the target state is no longer needed. The results of the experiments show that the tracking performance of the free-clustering optimal particle PHD filter is superior to the traditional particle PHD filter.