Multispectral multi-object tracking with enhanced motion cues for advanced unmanned maritime surveillance
Xiangen Bai, Zhijie Zhang, Xiaofeng Xu, Zhexin Xie, Luyuan Zou, Yahong Zhu · Ships and Offshore Structures · 2025
Unmanned intelligent maritime platforms, such as unmanned surface vehicles (USVs) and unmanned aerial vehicles (UAVs), play an increasingly vital role in maritime surveillance. However, in complex ocean environments, multi-object tracking (MOT) faces significant challenges, including target ambiguity and platform-induced jitter. Existing MOT approaches are typically based on single-modality data, limiting their ability to fully exploit the complementary advantages of multi-sensor information. This paper proposes a multispectral multi-object tracking method tailored for unmanned maritime surveillance. The approach leverages a lightweight and efficient multispectral detector, incorporates a motion-cue-driven data association framework, and introduces an Uncertainty-Gated Cascade Matching (UGCM) module to improve association robustness. Additionally, an Observation-Centered Kalman Filter (OCKF) is employed for posterior state estimation. Extensive experiments demonstrate the effectiveness of the proposed method. It achieves 41.2% HOTA, 44.4% MOTA, and 40.8% IDF1 on the VT-Tiny-MOT dataset (UAV perspective), and 37.2%, 39.9%, and 38.2% respectively on the Jari-RGBT-2024 dataset (USV perspective). The model operates at 35.4 FPS with only 12.2M parameters, achieving a favorable balance between accuracy, efficiency, and model size, making it well-suited for deployment on resource-constrained maritime unmanned platforms.