Moving Target Tracking Based on Shadow with Unsupervised Deep Track in Video-SAR

The moving target is always defocused and shifted out of the scene in image sequences in video Synthetic Aperture Radar (video-SAR), which makes moving target tracking be a formidable challenge. In this paper, a new moving target tracking method based on shadow with unsupervised deep track (UDT) is proposed. Our track network is built on the Siamese network framework and is trained on a series of unlabeled SAR image sequences with unsupervised learning way. Meanwhile, we use the trajectory detection technology to get more accurate tracking results. Finally., the processing results of the real data are provided to verify the effectiveness of the proposed method.