Multiview Fusion With the Labeled Multi-Bernoulli Densities in the Network Without Feedback

This article addresses the multiview multisensor fusion problem: the limited fields of view (FoVs) of the agents in the network are partially/nonoverlapping, thereby inducing the discrepancy among the local multitarget densities. Besides, for labeled random finite sets (RFSs), the labels are locally generated by the agents, which causes label inconsistency among the local label spaces. In this article, a fusion method is proposed to solve the multiview fusion problem and the label inconsistency problem simultaneously. Specifically, an information complement method was proposed for appending the virtual LMB components to each local multitarget density, reducing the difference among the complemented local multitarget densities. This method is realized by solving an optimization problem based on the GCI divergence. However, this problem is hard to solve, thus, a greedy algorithm is applied to obtain a suboptimal solution. The upper bound of the gap between the suboptimal and optimal solutions is also given in the article, and a cutoff threshold of the greedy algorithm is proposed according to this upper bound. Then the joint label GCI (JL-GCI) is used to obtain the fused multitarget density among the complemented local multitarget densities. The simulation results demonstrate the effectiveness of the proposed method.