Estimating Satellite Orientation through Turbulence with Deep Learning

Jacob Lucas, The Boeing Company; Trent Kyono, The Boeing Company; Michael Werth, The Boeing Company; Nicole Gagnier, The Boeing Company; Zackary Endsley, The Boeing Company; Justin Fletcher, SMC/DirSP-G; Ian McQuaid, Air Force Maui Optics and Supercomputing Site

Keywords: Machine Learning, pose estimation

Abstract:

Determination of satellite orientation is critical for evaluating spacecraft health and characterizing motion. Ground based sensors are under utilized for this task due to the image degradation imposed by atmospheric turbulence. We propose a deep learning based approach to real-time pose estimation from ground based sensor images. We leverage an extensive collection of simulated imagery and atmospheric turbulence to both train a pose estimator and to characterize performance under a variety of real scenarios. In addition, we show that for deeply degraded conditions where pose estimator performance suffers, the application of an image recovery network can restore accuracy without impacting real-time status.

Date of Conference: September 15-18, 2020

Track: Machine Learning for SSA Applications

View Paper