Space Target Tracking by Variance Detection

Detecting and tracking small targets in the aerospace is both a significant and difficult issue for the satellite tracking system. It is fundamentally challenging due to the existence of strong noise in captured images, the few characteristics of spot-like target, the simultaneous multiple targets, the real-time requirements, and so on. To address these challenges, we respectively design the detector and tracker who coordinate with each other. We formulate the detection stage as a two-step problem, which integrates the variance vector detection and 2th variance detection. The pixels of images are projected to the variance subspace of two-dimension. In the first step, the candidate targets are extracted with the optimal threshold achieved by K-means and proposed Weighted Maximum Right Probability, namely WMRP. In the second step, the true targets are checked out by adopting the proposed 2th variance feature and multiscale threshold. In the tracking stage, the Markov based dynamic model forecasts the probable area of the target in next time which is presented to the variance detector. Then the real location of the target estimated by the detector is transmitted to the tracker to generate the next probable area. Experiments demonstrate the proposed two-step framework can efficiently and rapidly detect the small multiple targets, and the cooperative working of the detector and tracker can satisfy the conventional application of space target tracking.