VideoSAR Target Tracking Algorithm by Integrating Shadow Detection Network and Global Nearest Neighbour Association

Abstract:
Video Synthetic Aperture Radar (VideoSAR) is a new radar system that can achieve continuous high resolution and high frame rate imaging by using the terahertz band electromagnetic waves as the working carrier. The target’s motion causes its echo signal displayed and defocused, while the shadows appear at the actual position of the target. In this paper, we propose a shadow based VideoSAR target tracking algorithm to detect and track the ground moving target via the VideoSAR image sequences. Considering that the shadow was low contrast and indistinguishable to the background, we propose a CenterNet based shadow detection network to detect the target shadow at VideoSAR frames. And then the Kalman filter (KF) and global nearest neighbour (GNN) association was utilized to associate the inter-frame detection results by minimize the distance between the target motion prediction and intra-frame detections. The experiments on the real VideoSAR data shows that the proposed tracking algorithm achieves higher qualitative tracking trajectories and higher multitarget tracking accuracy (MOTA) than the others.