A Density Based Unsupervised Learning Method for Radar Target Route Estimation

Tracking radar targets generally relies on kinematic characteristics from signal processing, since there is no specific priori knowledge of routes information in an area. However, it is a passive way that only using kinematic characteristics to tracking targets, and cannot further predict and analyze the target motivation in advance, losing important information. In practice, targets moving in the sea/air are following assigned routes due to the constraints of fuel, weather and expense. Aim at this problem, in this paper, we propose a method Density Based Route Extraction (DBRE) to extract routes from accumulated data without any information in advance. It finds hot regions where targets usually appears, labels noise data, and then extracts routes. Through comprehensive and comparative experiments, the performance of DBRE on route extraction is validated, especially in the aspects of route coverage and correctness, proving that our method provides an effective way to construct route network in an area when no extra information is available.