Efficient geometric matching with polar bounds for aligning star field images

Abstract

For many autonomous satellites, maintaining current knowledge of their location and pose is critical to mission success. Whilst there exist mature solutions for positioning and attitude estimation in space, computer vision approaches are emerging as viable and cheaper alternatives. At the core of a vision-based satellite motion estimation system is registering the successive images taken in space. This is a challenging task due to the lack of visually salient features in star eld images. In this research, we propose a novel technique for registering star eld images. We formulate the task as a geometric matching optimisation problem, which is solved using branch-and-bound. Our main contribution lies in devising a novel polar bounding function that provides tighter bounds than existing methods. Our bounding function is also easy to evaluate, requiring no more computations than available bounding functions. As demonstrated on actual space images, both factors contribute to speed up geometric matching by a large degree. The proposed technique further illustrates the promise of vision-based sensors for space missions.