Machine Vision Techniques for Planetary Terminal Descent Hazard Avoidance and Landmark Tracking

Abstract

This paper summarizes recent work at Martin Marietta on machine vision techniques to autonomously guide an unmanned spacecraft to a safe landing on a planet such as Mars. The work has concentrated on three areas: (1) monocular techniques to detect rough surface hazards such as boulder fields, (2) optical flow or "shape-from-motion" techniques to determine the three dimensional structure of the terrain and detect hazards such as steep slopes, and (3) recognition of landmarks to reduce navigation errors and achieve a more precise landing. Some of the landmark recognition and texture analysis techniques have been integrated into a real-time hardware simulation using a 6 degree-of-freedom robotic device and scaled terrain boards. This integration resulted in development of an autonomous guidance, navigation and control system with closed-loop image processing.