Multipath Extended Target Tracking with Labeled Random Finite Sets

TL;DR

This work introduces the MPET-GLMB filter, a unified Bayesian approach for tracking multiple extended targets in dynamic multipath radar environments. By integrating labeled RFS theory with extended target models (GGIW for objects and uniform sticks for reflectors) and a path-aware likelihood, the method jointly resolves target existence, measurement partitioning, and path associations to estimate trajectories for both objects and reflectors. A Gibbs-sampling-based joint prediction/update and a measurement-driven adaptive birth model enable tractable, robust performance, demonstrated to outperform state-of-the-art MP-ET-PHD and GLMB approaches on simulated and real automotive radar data. The results show superior state estimation accuracy and trajectory continuity in challenging multipath scenarios, with practical implications for autonomous driving and radar sensing in cluttered environments.

Abstract

High-resolution radar sensors are critical for autonomous systems but pose significant challenges to traditional tracking algorithms due to the generation of multiple measurements per object and the presence of multipath effects. Existing solutions often rely on the point target assumption or treat multipath measurements as clutter, whereas current extended target trackers often lack the capability to maintain trajectory continuity in complex multipath environments. To address these limitations, this paper proposes the multipath extended target generalized labeled multi-Bernoulli (MPET-GLMB) filter. A unified Bayesian framework based on labeled random finite set theory is derived to jointly model target existence, measurement partitioning, and the association between measurements, targets, and propagation paths. This formulation enables simultaneous trajectory estimation for both targets and reflectors without requiring heuristic post-processing. To enhance computational efficiency, a joint prediction and update implementation based on Gibbs sampling is developed. Furthermore, a measurement-driven adaptive birth model is introduced to initialize tracks without prior knowledge of target positions. Experimental results from simulated scenarios and real-world automotive radar data demonstrate that the proposed filter outperforms state-of-the-art methods, achieving superior state estimation accuracy and robust trajectory maintenance in dynamic multipath environments.

Figures (8)

• Figure 1: Radar multipath propagation geometry.
• Figure 2: Simulated ground truth for Scenario 1. The red cross denotes the radar position. The blue thick line represents the reflector. Square markers show the object centroids at selected time steps, with ellipsoids denoting the object spatial extents. Objects are assigned different colors for identification.
• Figure 3: OSPA distance, OSPA$^{(2)}$ distance and cardinality estimation error averaged over 100 Monte Carlo runs for Scenario 1.
• Figure 4: Simulated ground truth for Scenario 2. The two larger objects in the lower region move from left to right, while the two smaller objects in the upper region move from right to left.
• Figure 5: All measurements in one Monte Carlo run for Scenario 2.