A collaborative GMPHD filter for fast multi-target tracking

Unmanned Aerial Vehicle (UAV) which installed radio frequency radar is utilized in many applications for accurately target tracking. The Gaussian mixture probability hypothesis density (GMPHD) filter is a powerful algorithm for target tracking with significant performance. But in the UAV application scenarios with dense targets and intensive clutters, high computational complexity becomes a serious problem for GMPHD algorithm. By considering the differences of dynamic evolution between the survival target and birth target, a collaborative Gaussian mixture PHD (CoGMPHD) filter for fast multi-target tracking used in UAV system is proposed. This algorithm strives to improve the systematic implementing efficiency as well as guaranteeing the tracking accuracy by dynamically partitioning the measurement set into two parts, survival and birth target measurement sets. Gaussian components are updated respectively in each set, and an interactive and collaborative mechanism between the survival Gaussian components and birth Gaussian components is constituted. Simulation results shows that the proposed CoGMPHD filter guarantee the tracking accuracy as well as decreasing the computational complexity.