Through-Wall Radar Multitarget Tracking With Social-Force-Augmented Labeled RFS and Adaptive Observation Fusion

Abstract

Multitarget tracking with through-wall radar (TWR) is severely degraded in realistic propagation scenarios by strong static clutter, sidelobe leakage, multipath scattering, and target crossings. These effects induce split detections, elevated false-alarm rates, and partial occlusion, leading to identity switches, ghost tracks, and trajectory fragmentation. This article proposes an enhanced labeled random finite set (RFS) multitarget tracking framework for TWR. While preserving the analytical structure of the $\delta$-generalized labeled multi-Bernoulli ($\delta$-GLMB) filter, a social force model is embedded into the labeled state transition to capture interactions among targets and between targets and obstacles and strengthen the motion prior. In the measurement update, an adaptive energy-weighted observation fusion mechanism is introduced to treat nearby split returns as alternative realizations of a single detection event by marginalizing an energy-adaptive intracluster selection, yielding a $\delta$-GLMB-compatible cluster-level likelihood-ratio update for robust association in cluttered TWR scenes. The resulting tracker improves label consistency in dense and cluttered scenes, suppresses multipath- and sidelobe-induced ghost tracks, and enhances measurement-to-track discrimination. Simulations and multitarget TWR experiments demonstrate substantial gains over representative baseline trackers in the optimal subpattern assignment (OSPA$^{(2)}$) metric, localization error, and cardinality error, together with markedly fewer identity switches, false tracks, and fragmented trajectories. The proposed framework retains the generality of labeled RFS filtering and can be extended to other radar modalities and complex sensing scenarios.