The GMPHD Filter for Swarm Target Tracking Based on Gamma Gaussian Processes
Xi Cao, Yunlian Tian, Yiru Lin, Yunfei Liang, Wei Yi · 2024
The swarm target involves numerous individuals with close proximity and uniform sizes, rendering their situation challenging to discern. Swarm target tracking (STT) aims to simultaneously estimate their shape, center state, cardinality (i.e., the number of individuals), and other parameters using a limited-resolution sensor. To estimate the arbitrary and time-varying shapes of multiple swarm targets, we model them as star-convex and use a Gaussian process (GP) approach to estimate them recursively. Additionally, we utilize amplitude information to construct hypotheses and the Gamma distribution to estimate the cardinality of the swarm target. On this basis, we propose the Gamma GP probability hypothesis density (Gamma-GP-PHD) filter for STT and develop its Gaussian mixture (GM) implementation using merged polar measurements with amplitude. Some implementation issues and strategies are discussed. The performance of the proposed filter is validated in a scenario with multiple swarm targets of various shapes, and the results show its superiority of shape and cardinality estimation.