Graph Feature Representation for Shadow-Assisted Moving Target Tracking in Video SAR

Recently, video synthetic aperture radar (video SAR) has drawn widespread attention due to its capability to monitor moving targets continuously. Tracking the moving targets in video SAR using the shadow information has been proven as a more effective method. However, the existing tracking methods process each target independently and ignore the interframe interactions. To deal with this issue and improve the tracking performance, we propose a graph feature representation algorithm for video SAR multitarget tracking (MTT) using the global topological information. Specifically, a directed graph is built for each detected shadow based on the neighbor spatial relations, where each node is the semantic features of the corresponding shadow and each edge is the relative position features with neighboring shadows. Subsequently, the detected shadows are associated with the tracking shadows according to the similarity of their graphs to achieve moving target tracking. Experimental results on the video SAR dataset validate that compared with the state-of-the-art (SOTA) tracking algorithms, our algorithm has higher tracking accuracy and lower identity (ID) switching rate.