MSc Project: Development of an Open-Source Computer-Vision Based Orientation System for Cubesats
Aug 20, 2019
Work carried out by University of Strathclyde MSc Advanced Mechanical Engineering with Aerospace student, Daniel Hingston over the summer of 2019
Abstract
There is much potential for the development of orientation systems for CubeSats. This project covers the creation of an experimental open source computer vision orientation system, specifically for use on CubeSats. A critical literature review studied existing orientation systems, neural network uses in astronomy, and specific tools and approaches that would be applicable. The OpenCV library functionality for cascade classifier training was employed, and HAAR and LBP feature types were investigated. Different training and detection settings were studied to tune the detection system for the best performance. A final set of trained cascades was produced for the northern celestial hemisphere, and tested as a complete set. Further experiments were performed to investigate the chief causes of error. It was found that the primary area for improvement is to increase the rotationally invariant performance of detection, which may be an inherent weakness of an approach utilising HAAR and LBP feature types. Future work is suggested to adopt the successful aspects of this project, develop the idea further, and progress towards a deployable experimental system.