Gaussian Mixture Particle Jump-Markov-CPHD Fusion for Multitarget Tracking Using Sensors With Limited Views
Kai Da, Tiancheng Li, Yongfeng Zhu, Qiang Fu
Abstract
In this article, we propose a multisensor cardinalized probability density hypothesis (CPHD) filter for tracking an unknown number of targets that may maneuver over time by using a sensor network with partially overlapping fields of views (PO-FoVs). We develop a novel, Gaussian mixture particle (GMP) implementation of the jump-Markov CPHD filter to deal with highly non-linear/non-Gaussian models and target maneuvers. The concepts of zero-forcing and zero-avoiding originally used in density approximation are introduced to elucidate a key difference between geometric and arithmetic averaging approaches, which are extended for joint target-state and mode fusion with regard to each PO-FoV for which distributed flooding is used for internode communication. The resulting GMP-JMCPHD fusion algorithm comprises three FoV-oriented steps: splitting, fusion, and merging. Simulations are provided to demonstrate the effectiveness of the proposed approaches.