Robust distributed fusion with trajectory random finite sets.

• This paper concentrates on the problem of distributed Multi-sensor Multi-target Tracking (DMMT) in the trajectory random finite set filtering framework. The trajectory probability hypothesis density (PHD) filter and the trajectory cardinality PHD are performed locally in different sensors for DMMT. • However, it is not convenient to use fusion techniques for the sets of targets directly to sets of trajectories. The reason is that optimal assignment among different trajectories in different sensors may change across time, and we should account for this in a fusion algorithm. • We pursue the trajectory association algorithm based on Clustering and Linear programming, which is one of the key ingredients of trajectories fusion. The trajectory association solves a joint optimization problem such that the cost of localization and switching over all time steps is minimized. • This paper proposes two cardinality consensus algorithms to fuse unassociated and covariance intersection (CI) fused trajectories. We also calculate the crucial parameters in CI fusion via unitary decomposition and eigenvalue decomposition. • The filters of DMMT developed by this paper can widely accommodate cardinality inconsistency between local multi-trajectory, limited computing and communication resources of nodes, various fields of view. This paper concentrates on the distributed Multi-sensor Multi-target Tracking (DMMT) in the trajectory random finite set (TRFS) filtering framework. The trajectory probability hypothesis density (TPHD) filter and the trajectory cardinality PHD (TCPHD) are performed locally in different sensors for DMMT. Our analysis shows that standard generalized covariance intersection (GCI) and weighted arithmetic average (WAA) fusion with the posterior multitrajectory densities of the TPHD and TCPHD filters rely heavily on the proper association of trajectories estimated from various sensor nodes. Then, inspired by the analysis, we develop two novel distributed fusion algorithms for the Gaussian mixture TPHD and TCPHD filters. First, we obtain the associated and unassociated trajectories at different sensors via clustering and Linear programming (LP). Second, the associated trajectories are fused via covariance intersection (CI) parallelly. Third, two principle cardinality consensus algorithms are proposed to fuse unassociated and CI fused trajectories. Moreover, two analytical expressions, which are crucial for fusion weight in CI fusion, are provided. Finally, numerical experiments demonstrate the efficacy of our proposed approaches in computational efficiency and accuracy, compared with the state-of-the-art solutions in challenging scenarios. [ABSTRACT FROM AUTHOR]