Track-before-detect based multiple vehicle tracking using two spatially separated linear microphone arrays

Abstract

Multi-vehicle acoustic tracking is a complex task facing multiple challenges, including overlapping measurements among targets and signal blockage due to occlusion. These problems are exacerbated by weak target responses over long propagation distances and multiple spatially separated acoustic source components. To address these challenges, this paper proposes a random finite set (RFS)-based multi-vehicle acoustic tracking method for two-dimensional vehicle position estimation using acoustic beamforming measurements from a spatially distributed linear microphone array system. Specifically, a track-before-detect (TBD)-based labeled multi-Bernoulli (LMB) filter with an interference-normalization model is developed, in which multi-bin angular sound intensity information is incorporated into the multi-target tracking framework. The proposed method has been evaluated on real-world acoustic data collected in four challenging scenarios involving multiple vehicles with varying motion patterns. In addition, a bird’s-eye-view drone video was captured to obtain ground-truth trajectories through manual annotation. The tracking performance was compared with prominent methods, such as the traditional LMB filter and the probability hypothesis density (PHD) filter, using the OSPA and OSPA ( 2 ) metrics. Experimental results demonstrate that the proposed method achieves lower localization and cardinality errors and provides more stable multi-target tracking performance in challenging scenarios.