Improving Labeled Random Finite-Set-Based Filtering With Limited Fields of View

The challenge of multitarget tracking (MTT) with sensor networks with limited fields of views (FOVs) is seeing increasing interest in the MTT community. Most work on this problem to date has focused on single hypothesis filters, which suffer greater information loss than multiple hypothesis filters. However, multiple hypothesis filters tend to experience runaway growth in computational complexity in the presence of significant FOV limitations. This article presents a modification to the generalized labeled multi-Bernoulli filter that enables tractable multiple hypothesis MTT under such conditions. This modification is shown to generalize to a variety of labeled random finite-set-based filters. We apply various filters with different degrees of modification to a simulated MTT scenario with two limited-FOV sensor networks: one with significant separation between sensor FOVs and one with significant overlaps. Our modification is shown to provide improvements in accuracy and computational efficiency in both cases, with greater overall impact in the case with separated FOVs.