Network-Wide Consistent Fusion for Distributed Multitarget Tracking Under Limited and Occluded Fields of View

Abstract

Distributed multitarget tracking (DMTT) enables multiple sensor nodes with heterogeneous and limited fields of view (FoVs) to cooperatively estimate target trajectories through network-wide fusion. However, when FoVs are occluded, local tracks may become fragmented and inconsistently labeled, and these local identity errors can propagate through distributed fusion. To address these issues, this article proposes consistent fusion by associated groups algorithm with a generated label mechanism (CFAG-GLM), a track-level network-wide fusion framework that integrates associated-group fusion with a generated label mechanism. Internode track association and intragroup state fusion produce consistent fused estimates across nodes, while a unified global-label update maintains persistent fused identities across scans. Within this process, a rebirth relation based on temporal and kinematic consistency reconnects reinitialized local tracks to existing global identities after occlusion-induced interruptions. Comparisons with density-level and track-level fusion baselines, together with a no-rebirth ablation, show that CFAG-GLM preserves network-wide consistency while reducing trajectory fragmentation and identity discontinuities in the tested occluded-FoV scenarios. In the two occluded-FoV scenarios, CFAG-GLM achieves the lowest OSPA ${}^{\text {(}{2}\text {)}}$ and LP switching errors among the compared methods while retaining runtime comparable to the track-level baselines. Relative to CFAG, it reduces the occlusion-induced OSPA ${}^{\text {(}{2}\text {)}}$ degradation by 69.2% and 65.7%, respectively, and the corresponding LP total-error degradation by 42.9% and 43.3%, respectively.