Attention-based Multi-Target Shadow Tracking for Video SAR

Achieving high-precision and reliable tracking in complex target motion, low shadow quality, occlusion, blurring, and other scenarios is the main problem of multi-target tracking in Video SAR. This paper presents an enhanced multi-target tracking algorithm that adds attention mechanisms to improve target feature extraction and matching, performs preliminary object detection and tracking based on the TraDes model, and first preprocesses the image sequence for enhancement and denoising. With the model focused on the region of interest and improved inter-frame similarity, the improved algorithm shows higher tracking accuracy and robustness, effectively improving multi-target tracking performance under complex conditions like limited representation ability and noise interference caused by unclear shadow appearance features. It provides a lot of advantages over conventional techniques. This offers fresh concepts for the advancement of multi-target tracking video SAR technology. According to the experimental results using the Sandia National Laboratory (SNL) dataset, our technique outperformed JDE (33.56%), FairMOT (39.18%), and CenterTrack (40.72%) with the highest MOTA score (46.33%) in the video SAR test sequence, which was 12.77%, 7.15%, and 5.61% higher.