Visual Tracking Methods for Improved Sequential Image-Based Object Detection

This paper applies the random finite set based multi-Bernoulli filter with a detectionless likelihood function to frame-to-frame tracking of space objects observed in electro-optical imagery for space domain awareness applications. First, this paper reviews multi-Bernoulli filters applied to frame-to-frame tracking, image statistics, and matched filters. A likelihood function for space-based imagery is analyzed in comparison to the previously used likelihood function. A birth model is proposed that better models potential space objects using observer characteristics and object dynamics. In simulation, the final algorithm is able to perform completely uncued detection down to a total photometric signal-to-noise ratio of 5.6 and a per-pixel signal-to-noise ratio of 1.5. Promising results are shown for a total photometric signal-to-noise ratio of 3.35 and per-pixel signal-to-noise ratio of 0.7. The algorithm is also applied to empirical data, which involves tracking of low signal-to-noise ratio geostationary objects in images taken with a 0.5 m Raven-class telescope.