Author: Fengliang Xu

Publisher:

Published: 2004

Total Pages:

ISBN-13:

Abstract: In the exploration of an extraterrestrial environment such as Mars, orbital data can not replace the role of landers and rovers, which can provide a close up and inside view. Orbital mapping can not compete with ground-level mapping in resolution, precision, and speed. This dissertation addresses two tasks related to robotic extraterrestrial exploration: mapping and rover localization. Image registration is also discussed as an important aspect for both of them. Image registration is classified into three sub-categories: intra-stereo, inter-stereo, and cross-site. In the intra-stereo registration, interest point-based registration and verification by parallax continuity in the principal direction are proposed. Two other techniques, constrained dynamic programming search for far range matching and Markov Random Field based registration for big terrain variation, are explored as possible improvements. For inter-stereo, a so-called "shift-and-vote" algorithm are proposed. These techniques provide a basis for automation of image registration. Cross-site registration is a separate area of research, and it will not be covered in this dissertation. Creating using rover ground images mainly involves the generation of Digital Terrain Model and orthomap. A dual polynomial model is proposed for interpolation of the DTM, using Kriging in the close range and TIN in the far range. To generate a uniformly illuminated orthomap from the DTM, a least-squares-based automatic intensity balancing method is proposed. Finally a seamless orthomap is constructed by a split-and-merge technique. Rover localization has three stages, all of which use a least-squares adjustment procedure: (1) an initial localization by adjustment over features common to rover images and orbital images, (2) an adjustment of image pointing angles at a single site through inter and intra-stereo tie points, and (3) an adjustment of the rover traverse through manual cross-site tie points. The first stage is based on adjustment of observation angles of features. The second stage and third stage are based on bundle-adjustment. Automation in rover localization includes automatic intra/inter-stereo tie point selection, computer-assisted cross-site tie point selection, and automatic verification of accuracy. These methods are demonstrated with datasets from 2004 Mars Exploration Rover mission as well as several past missions and simulations.