A Probability Hypothesis Density Filter with Target Competitive Re-Update Scheme for Tracking Close Proximity Targets

The Gaussian mixture probability hypothesis density (GM-PHD) provides a closed-form solution to the probability hypothesis density (PHD) filter, which is a promising method to track multiple targets in clutter scenarios. However, the PHD filter exhibits inferior performance in terms of target states and their number in close proximity target scenarios. By applying a novel target competitive re-update scheme for the GM-PHD filter, a multiple closely spaced target GM-PHD (MCST-GM-PHD) filter is proposed to track multiple targets in close proximity. Based on the previous state estimates of targets, the target competitive re-update scheme is utilized to identify the most possible measurement originated from each target among the latest received measurements. With the correct target-originated measurements, the closely spaced targets can be correctly re-updated using its own true measurements in the updated step. Numerical simulations are performed in challenging scenarios involving targets with crossing and paralleling tracks, where the proposed MCST-GM-PHD filter is able to effectively estimate the state and cardinality of close proximity targets and is significantly better than the related multi-target PHD algorithms.