OptiPMB: Enhancing 3D Multi-Object Tracking with Optimized Poisson Multi-Bernoulli Filtering

TL;DR

OptiPMB tackles 3D multi-object tracking in autonomous driving by leveraging an optimized PMB filter within a tracking-by-detection framework. It introduces a measurement-driven hybrid adaptive birth model, adaptive detection probabilities, and density-pruning/track-extraction optimizations to strengthen RFS-based tracking. The approach achieves state-of-the-art performance on nuScenes and KITTI, notably attaining AMOTA improvements (e.g., 0.767 AMOTA on nuScenes with LargeKernel3D) and enhanced ID maintenance. This work establishes a practical and principled benchmark for PMB/RFS-based online 3D MOT and offers design insights for robust birth, data association, and track management in cluttered driving scenes.

Abstract

Accurate 3D multi-object tracking (MOT) is crucial for autonomous driving, as it enables robust perception, navigation, and planning in complex environments. While deep learning-based solutions have demonstrated impressive 3D MOT performance, model-based approaches remain appealing for their simplicity, interpretability, and data efficiency. Conventional model-based trackers typically rely on random vector-based Bayesian filters within the tracking-by-detection (TBD) framework but face limitations due to heuristic data association and track management schemes. In contrast, random finite set (RFS)-based Bayesian filtering handles object birth, survival, and death in a theoretically sound manner, facilitating interpretability and parameter tuning. In this paper, we present OptiPMB, a novel RFS-based 3D MOT method that employs an optimized Poisson multi-Bernoulli (PMB) filter while incorporating several key innovative designs within the TBD framework. Specifically, we propose a measurement-driven hybrid adaptive birth model for improved track initialization, employ adaptive detection probability parameters to effectively maintain tracks for occluded objects, and optimize density pruning and track extraction modules to further enhance overall tracking performance. Extensive evaluations on nuScenes and KITTI datasets show that OptiPMB achieves superior tracking accuracy compared with state-of-the-art methods, thereby establishing a new benchmark for model-based 3D MOT and offering valuable insights for future research on RFS-based trackers in autonomous driving.

Figures (5)

• Figure 1: Illustration of the RFS-based 3D MOT system model. The top section displays bounding boxes in different colors, representing tracks with distinct IDs estimated by the RFS-based tracker.
• Figure 2: Overall pipeline of the proposed OptiPMB tracker. Compared to the previous state-of-the-art PMB filter-based 3D MOT method GNN_PMB, the innovative differences and improvements of OptiPMB are highlighted in distinct colors. Yellow denotes adaptive designs to improve the robustness of the tracker (see Section \ref{['Adaptive Designs']} for details). Green denotes algorithm modules optimized for better tracking performance (see Section \ref{['Optimizations']} for details).
• Figure 3: Cost matrix and local data association hypotheses. The misdetection, detection, and first-time detection hypotheses are highlighted in orange, green, and blue, respectively. Cost matrix entries with infinity values represent impossible association hypotheses.
• Figure 4: Qualitative comparison among our proposed OptiPMB, Fast-Poly Fast_Poly, GNN-PMB GNN_PMB, and ShaSTA ShaSTA on the nuScenes validation set (Scene-0095). Black solid boxes denote ground truth objects. Dashed boxes represent estimated objects, with distinct colors indicating different IDs. Dashed colored lines depict object trajectories. ID switch errors are highlighted by red arrows.
• Figure 5: Impact of key OptiPMB parameters on tracking accuracy for cars in the nuScenes validation set with CenterPoint CenterPoint detections. Selected parameters are used to report the results in Table \ref{['nuScenes val performance']}.