The extraction of faint space objects with unknown orbits from digital imagery is considered to be a problem. An image is recorded using a wide-field of view (FOV) electro-optics sensor operating in stars-tracking mode. To solve this problem an approach based on the detection of fragments of tracks with further grouping is considered. At the same time the problem of stars detection is solved. Both fragments of tracks and stars are detected within a sliding spatial window with the classification of "star / fragment of track". Signal accumulation along a hypothetical track is restricted by the spatial window. The total number of extracted track fragments to be grouped is much less than the total number of hot pixels to be associated into the tracks if the traditional approach is used, based on a low signal threshold.