Time-space domain assignment for generalized covariance intersection fusion with labeled multitarget densities

Abstract

This paper addresses the generalized covariance intersection (GCI) fusion for labeled random finite sets, avoiding the label inconsistency sensitivity problem and the information loss for target identities. In this paper, the label assignment is considered from the perspective of space and time. First, to solve the label inconsistency sensitivity problem, the joint label space for the support of fused labeled random finite sets is proposed to represent the target identity and the assignments of labels of different local label spaces, followed by obtaining the multitarget density over the joint label space via the GCI fusion. To avoid the information loss for target identities so as to improve the performance of the track continuity, we utilize all the label information to establish the assignment of targets across the time steps. Specifically, the identity distribution for each target is extracted from the global multitarget density and then utilized for computing the Cauchy-Schwarz divergence (CSD) of targets across the time steps. The assignment between the target identities across the time steps is obtained by formulating a linear assignment problem, which can be effectively solved by the Hungarian algorithm. Simulation results exhibit the effectiveness of the proposed approach in challenging scenarios.