Abstract

Dead spacecraft and burnt-out rocket bodies and the pieces of spacecraft and rocket that fragments from old spacecraft collisions make are getting more and more crowded in orbit around the planet. The smallest and faintest of these objects is hard to detect with surveys based on radar or the classical pipelines of telescopes. In this article, we will explore how the current state of the art neural-network vision technologies could be leveraged to address that challenge and outline a five-stage processing flow: Calibration, simulated-data augmentation, multi-scale feature learning, frame-to-frame detection and tracking. The recall, false alarms, etc. and computing cost test plans are also described. The chain is a proposal and hasn’t been tested experimentally, as there is no common set of annotated real debris images.

Keywords

How to Cite This Article

Rights