On Lunar On-Orbit Vision-Based Navigation: Terrain Mapping, Feature Tracking Driven EKF

Abstract

This paper presents an approach to vision guided, feature-based navigation that integrates absolute navigation concepts involving image feature extraction and matching to a database, with relative navigation including feature flow and tracking. The approach uses an extended Kalman filter (EKF) for vehicle navigation to integrate the two methodologies and thus capture the advantages of both techniques. This methodology was specifically developed for navigation in lunar orbiting or landing applications. Crater fiducial markers from the lunar surface are identified in an image and associated to a database; these are tracked across the image plane as the lunar lander moves in its orbit. In addition to the vision sensor, an altimeter is used to supply altitude measurements above the surface of the moon. The EKF presented here estimates the inertial vehicle position and velocity states by tracking the fiducials, whose inertial coordinates are now known, across the image plane. The accuracy improvements of this approach are substantial compared to either the absolute navigation pose estimation methods or the relative navigation optical flow methods when used alone in the same scenarios. The dominant computational overhead in any of these methods is in acquiring and processing the images for the features. Therefore, there is little additional processing load associated with this method that either of the other vision-based techniques do not require. Images used in this research to develop and analyze the algorithms were obtained from NASA’s Clementine lunar mapping mission.