Siam-Sort: Multi-Target Tracking in Video SAR Based on Tracking by Detection and Siamese Network

Shadows are widely used in the tracking of moving targets by video synthetic aperture radar (video SAR). However, they always appear groups SAR images. In such cases, track effects produced existing single-target methods no longer satisfactory. To this end, an effective way to obtain capability multiple target (MTT) is urgent demand. Note that detection (TBD) for MTT optical images has achieved great success. TBD cannot be utilized directly. The reasons shadows quite different from image than as time-varying and their pixel sizes small. aforementioned characteristics make hard detect process lead numerous matching errors data association process, which greatly affects final performance. Aiming at above two problems, paper, we propose a method based on Siamese network. Specifically, improve accuracy, multi-scale Faster-RCNN first proposed targets. Meanwhile, dimension clusters accelerate convergence speed model training well better network weights. Then, SiamNet reduce errors. Finally, apply Kalman filter update results. experimental results real datasets demonstrate outperforms other state-of-art methods, ablation experiment verifies effectiveness SimaNet.