Efficient Distributed PHD Filtering via Sampling Clustering

This paper proposes a novel distributed approach to address the multi-target tracking problem in complex scenarios. We confront two challenges: one is the absence of prior knowledge in sensor statistics, including measurement noise, target detection probabilities, and clutter rates; the other is the need to improve both tracking accuracy and computational efficiency simultaneously. The proposed approach introduces an enhanced measurement preprocessing framework that transforms raw multi-sensor data into synthetic pseudo-likelihood measurements via a sampling clustering (SC) algorithm. First, the average distance derived from a distance matrix is utilized to identify measurement subsets generated by potential targets. Subsequently, non-empty and overlapping measurement subsets are adaptively merged to form distinct clusters, while the cluster centroids are determined based on the inferred target cardinality. These cluster centroids, acting as proxy measurements, are then integrated into a reformulated probability hypothesis density (PHD) filter update equation to estimate target states. Simulation results demonstrate that the proposed approach achieves superior tracking accuracy and computational efficiency compared to advanced distributed PHD filters, particularly in scenarios with unknown sensor statistics and dense clutter.