Multi-Task Based Marine Radar Tracking Network

Marine radar is necessary in the ship monitoring system. Multi-target tracking is the basis for automatically monitoring, the tracking methods should have both real-time efficiency and accuracy. However, many traditional methods do not perfectly balance computation cost and precision, especially when they are suffered from the clutters and other interference. Even if some methods simplify the traditional tracking process with deep learning, the efficiency of the entire tracking flow is still far from real-time. In this paper, we proposed a practical two-stage tracking-by-detection scheme consisting of a multi-task network and a simple data association method. The network produces the detection results and target appearance features at the same time, which can reduce the computational cost. And we enhance the radar images according to the properties of radar images to improve the accuracy and precision of detection results and appearance features. We conduct the performance assessment experiments on real data acquired by an X-band marine radar, and the results can demonstrate our method is practical and efficient.