——— 
AUTOMATION IN MARS LANDING-SITE MAPPING AND ROVER LOCALIZATION 
Fengliang Xu 
Dept. of Civil & Environmental Engineering & Geodetic Science, The Ohio State University 
470 Hitchcock Hall, 2070 Neil Ave., Columbus, OH 43210 - xu.101@osu.edu 
Commission VI, WG VI/4 
KEY WORDS: Extraterrestrial, Planetary, Exploration, Close Range Photogrammetry, Mapping, Block Adjustment, Automation, 
DEM/DTM, Triangulation, Rectification 
ABSTRACT: 
Our project aims to automate Mars mapping and localization using robotic stereo and descent imagery. Sterco vision is a well- 
studied domain. However, most efforts aim only at a general scene; little work has been done toward a natural, extraterrestrial 
environment through consideration of its special geometry and features. Our methodology utilized the properties of piece-wise 
continuity of natural scene and monotonously decreasing parallax between horizontal-looking stereo cameras. In outline, our 
automation process for processing robotic stereo imagery is: 1) interest points are extracted as features and matched between intra- 
and inter-stereo images, 2) tie points selection, 3) images with various illumination are balanced through use of tie points, 4) DEMs 
interpolation from 3-D interest points using Kriging and TIN, 5) orthophotos generation with DEM, and 6) landmarks (i.e. rocks) are 
extracted and occlusions are marked. Then, with the help of orthophotos, landmarks from different locations can be identified. 
Finally, robot localization is accomplished through use of rigid transformation and bundle adjustment of matched landmarks. For 
descent imagery, lower-level images are resampled and registered to higher-level images. Elevation is then estimated from multiple 
observations. This methodology has been used in the NASA 2003 Mars Exploration Rover Mission (MER) for precise robot 
navigation and mapping in support of the MER 2003 science and engineering team objectives. 
1. INTRODUCTION 
1.4 Background 
The current hotspot of extraterrestrial exploration is Mars. 
Compared with other planets of this solar system, Mars is the 
one most similar to Earth, so it becomes the first stop in 
searching for extraterrestrial life. Water and life are closely 
related, and the search of water evidence on Mars is an 
important task in Mars exploration. In 2004, two twin rovers, 
Spirit and Opportunity, arrived at the equator of Mars, one on 
each side, and began their long journey to check rocks and 
craters for traces of water. 
The distance between Mars and Earth is 5.57—40.13x 107 km; it 
takes a radio signal around 20 minutes to finish a one-way trip. 
This delay, plus the limited window of communications 
between Mars and Earth (as the rovers use solar panels for 
power), makes it very hard to handle the rover directly from the 
Earth. Autonomous navigation is the most feasible way to make 
efficient use of the Mars rovers. Currently, instructions, 
including target location and approximate route, are sent to the 
rovers; rovers then use hazard avoidance techniques and 
navigation instruments to approach their target. Route planning 
and localization need detailed higher precision and higher 
resolution maps that can not be provided through satellite 
imagery. 
The MER rovers are equipped with four of stereo cameras: 
Navcam, Pancam, front Hazcam, and rear Hazcam. Navcam 
parameters are: 1024x1024 pixels, 0.82 mrad/pixel, 45° FOV, 
and 15cm baseline (Maki et al., 2003). Pancam parameters are: 
1024x1024 pixels, 0.27 mrad/pixel, 16? FOV, and 25cm 
baseline (Bell et al, 2003). The valid measurement range (1m 
distance measurement error with mismatch level at 1/3 pixel in 
parallax) is around 27m for the Navcam and around 52m for the 
Pancam. Thus the Navcam is used for close-range mapping and 
the Pancam is used for long-range mapping. 
Each rover takes photos at different looking angles. Camera 
rotations around the mast (azimuth) and around the camera bar 
(tilt) are recorded and serve as the initial orientation parameters. 
These photos can form a panorama, which normally consists of 
10 Navcam pairs or 27 Pancam pairs. The overlap between 
neighboring image pairs (inter-stereo) is around ten percent. 
Overlap between left and right images of a pair (intra-stereo) is 
around ninety percent for Navcam and seventy percent for 
Pancam. These images can be linked through tie-points. 
1.2 Brief review 
Stereo vision is a well-studied problem. Existing methods can 
be classified into local methods, global method, and occlusion 
detection (Brown, 2003), or into three steps: matching-cost 
calculation, aggregation, and optimization (Scharstein, 2002). 
Most methods are aimed at making a parallax image for the 
overlapping area of two images in a general scene scenario. The 
global methods include dynamic programming methods (Ohta, 
1985), which consider only constraints in the scanline direction, 
and graph cut methods (Roy, 1998), which apply constraints in 
both the scanline and the inter-scanline directions. The first 
method is limited; the second gives better result, but is very 
time consuming. 
For robot navigatiori, there are both local and global methods of 
stereo vision, as well as active vision (Desouza, 2002; Jensfelt, 
2001; Leonard, 1991). Most of these are used for indoor 
applications (Kriegman, 1989); some are used for unstructured 
1042 
PS TA o SA ftr | = 
junk 
px 
in 
in 
an 
si 
ur 
to 
bu 
Pl 
an 
to 
We 
Gt 
po 
the 
Th 
pai 
def 
lox