SeaReTrack: Motion-compensated multi-object tracking with boundary-aware identity recovery for USV maritime perception

Reliable multi-object tracking is essential for unmanned surface vehicles (USVs) to maintain continuous awareness of surrounding vessels and water-surface objects for downstream navigation and collision-risk assessment. However, RGB-based maritime tracking remains challenging because wave-induced camera motion, unstable detections, reflections, low illumination, and frequent target disappearance or re-entry can severely disrupt trajectory continuity and cause identity fragmentation. To address these practical challenges, this paper presents SeaReTrack, an online multi-object tracking framework tailored to USV maritime perception. SeaReTrack combines motion-compensated reliable association for stable frame-to-frame matching, lightweight HSV-based appearance verification as complementary identity evidence, and lifecycle-managed boundary-aware identity recovery for reconnecting historical identities after prolonged disappearance or image-boundary re-entry. The recovery stage separates routine active-track association from uncertain identity reactivation and progressively considers recovery-eligible Active, Lost, and Archived trajectories. Experiments on the public USVTrack benchmark show that SeaReTrack achieves 64.35 HOTA and 80.09 IDF1 with 28 identity switches while retaining real-time end-to-end processing capability on both desktop and embedded platforms. Additional evaluations under rainy, low-light, and target re-entry conditions further demonstrate its identity-preservation capability in challenging maritime environments.