Abstract: On orbit optical observations are not affected by factors such as clouds, atmosphere and daylight, and are superior to traditional ground based measurements in terms of temporal and spatial range and cost of use. Based on the needs of on orbit detection and tracking of non cooperative space targets, the paper proposes a method for on orbit detection of space multi targets. This method first uses Delaunay triangulation matching to achieve stellar background differentiation, and then integrates the target photometric, morphological and kinematic features to perform trajectory association, and finally performs trajectory recognition to achieve robust detection of space targets. In the simulation data set of dense star field multi target observations and field target observation experiments, the multiple object tracking accuracy is greater than 95%, and it can effectively output multi target trajectories without relying on star catalogs for attitude solution, camera calibration parameters and target prior information. Compared with traditional algorithms, it reduces the amount of calculation and improves the onorbit detection capability of space targets.

Key words: uncertainty, time delay, non-cooperative targets in space, on-orbit detection and tracking, delaunay triangle matching, multi-feature fusion