A Probabilistic Approach of Space Objects Detection from Non-Resolved Optical Observation

Non-resolved optical space imagery is usually heavily noisy with many field stars clutter. Both phenomena may impair the success of detecting space objects. There is hence often a need for efficient and robust preprocessing techniques to filter the objects of interest.

In this paper, we propose a novel probabilistic approach for denoising, filtering and detecting space objects by using non-resolved optical images. In particular, as pixels of background and foreground in an image obey different probabilistic distributions, we propose a corresponding clustering algorithm to distinguish foreground objects from background noises. Furthermore, a near real-time finer classification for foreground objects is achieved by further exploring the metric on the filtered pixel set. Various modes (sidereal/stationary) and different types of objects (GEO/LEO) are unified into this general framework. We verify the effectiveness and robustness of our algorithm by detecting and filtering LEO and GEO objects in CCD telescopic sequential imageries under different experimental conditions, and exploit associated light curves for further characterization.