SeaCrossTrack: A Framework for Cooperative UAV–USV 3D Object Detection and Tracking in Maritime Environments

Unmanned aerial vehicles (UAVs) and unmanned surface vehicles (USVs) provide complementary maritime observations, but heterogeneous viewpoints, duplicate detections, and intermittent visibility complicate cooperative object tracking. This paper presents SeaCrossTrack, an object-level framework that integrates observation-quality-aware fusion, multidimensional trajectory association, and shared track memory in a common world coordinate frame. The fusion module retains selected observation geometry and supplies quality information to track updating and confirmation, while the association module combines position, velocity, heading, dimensions, and stored trajectory states. Experiments on the simulated SeaV2U-SCT dataset show that SeaCrossTrack achieves MOTA, IDF1, and HOTA scores of 85.42%, 78.19%, and 62.58%, respectively. Under native lifecycle settings, SeaCrossTrack improves MOTA, IDF1, and HOTA over NMS fusion followed by AB3DMOT by 0.68, 7.47, and 1.43 percentage points, respectively, with the largest improvement observed in IDF1. Under unified track lifecycle rules, SeaCrossTrack retains a 3.63-percentage-point IDF1 advantage over NMS + AB3DMOT. QCF achieves BEV and 3D mAP scores of 85.01% and 81.65% at IoU 0.5. The evaluated serial implementation requires 303.05 ms per frame for dual-platform detection, fusion, and tracking, corresponding to an estimated computational throughput of 3.30 FPS, excluding communication and synchronization delays. On the evaluated simulated sequences, SeaCrossTrack improves aggregate identity matching, with the largest gain in IDF1 and a tradeoff in local continuity reflected by higher IDS and fragmentation counts.