In the exploration of the planets of our solar system, images taken during a lander's descent to the surface of a planet provide a critical link between orbital images and surface images. The descent images not only allow us to locate the landing site in a global coordinate frame, but they also provide progressively higher-resolution maps for mission planning. This paper addresses the generation of depth maps from the descent images. Our approach has two steps, motion refinement and depth recovery. During motion refinement, we use an initial motion estimate in order to avoid the intrinsic motion ambiguity. The objective of the motion-refinement step is to adjust the motion parameters such that the reprojection error is minimized. The depth-recovery step correlates adjacent frames to match pixels for triangulation. Due to the descending motion, the conventional rectification process is replaced by a set of anti-aliasing image warpings corresponding to a set of virtual parallel planes. We demonstrate experimental results on synthetic and real descent images.

One goal for future Mars missions is to navigate a rover to science targets not visible to the rover, but seen in orbital or descent images. In order to support and improve long-range navigation capabilities, we generate 3D terrain maps using all available images, including surface images from the lander and/or rover, descent images from the lander, and orbital images from current and future Mars orbiters. The techniques used include wide-baseline stereo mapping for terrain distant from the rover, bundle adjustment for high-accuracy mapping of surface images, and structure-from-motion techniques for mapping using descent and orbital images. The terrain maps are compiled using a system for unifying multi-resolution models and integrating three-dimensional terrains.

Image-matching applications, such as tracking and stereo, commonly use the sum-of-squared-difference (SSD) measure to determine the best match. However, this measure is sensitive to outliers and is not robust to template variations. Alternative measures have also been proposed that are more robust to these issues. We improve upon these using a probabilistic formulation for image matching in terms of maximum-likelihood estimation that can be used for both edge template matching and gray-level image matching. This formulation generalizes previous edge-matching methods based on distance transforms. We apply the techniques to stereo matching and feature tracking. Uncertainty estimation techniques allow feature selection to be performed by choosing features that minimize the localization uncertainty.

High precision topographic information from all available data is critical to many landing site geological and engineering applications. Meanwhile, precise navigation and localization of the rover is important for its safety, engineering and scientific objectives when it traverses the Martian surface. In order to support future long-range rover missions (for example, 1km traverse in the planned 2003 MER mission), high precision landing site mapping and rover localization are desirable. At a landing site, for example Mars Pathfinder (MPF) landing site, the mapping accuracy using lander images depends on the distance from the lander to the mapped objects because the lander imager is stationary and the resolution becomes lower for areas far away from the lander. On the other hand, the rover images cover the landing site with a much higher and even resolution, especially in the rover operation area. Therefore, taking advantages of rover images will be very helpful for producing the best topographic information of the landing site and for rover navigation. Supported by a NASA/JPL sponsored project “Mars Rover Localization Using Descent and Rover Imagery”, we have developed algorithms and software for integrated bundle adjustment of orbiter, decent/lander, and rover images. At first, a three dimensional image network is built by linking all the orbital and ground-based images together using a vast amount of tie points identified and selected automatically or manually. Some distinctive features, such as mountain peaks and craters that can be observed from the orbital images, serve as relative controls for the adjustment. And then, the camera positions and attitudes of the rover images as well as all other images and 3D ground positions of all tie points can be obtained by a least squares adjustment. An incremental bundle adjustment model was also developed, which adjusts descent and rover images step by step, resulting in computational efficiency. An innovative approach has been investigated for automatic feature extraction and tie point selection based on interesting point filtering and image matching techniques. The preliminary experiment showed that the approach works well for stereo images at individual camera stations. Extension to cross-station and between descent/lander and rover images will be tested. Techniques of generating hierarchical DEM using descent and rover images were also developed for Mars landing site mapping and rover localization. Based on the improved image orientation parameters and DEM, orthophoto images and mosaics can also be produced. In order to verify our algorithm and software, two field test were conducted in April 1999 and May 2000, respectively, at Silver Lake, CA. Based on the data various rover localization experiments were carried out. The rover localization accuracy has reached about 1m for a traverse length of 1km from the landing center, using descent and rover images by an integrated or incremental adjustment. The root mean square (RMS) errors of the ground coordinates of checkpoints are 0.195m, 0.211m and 0.429m in X, Y, and Z directions, respectively. The experiment results also show that if there is no descent images available (like MER mission), it is still feasible to localize the rover using rover images only and achieve the similar accuracy with more efforts in optimal traverse design and image network generation. Currently, we are conducting bundle adjustment using MPF lander and rover images. It is expected that with the combined rover with lander images in one image network, the overall mapping and rover localization accuracy will be greatly improved. The latest result will be presented at the workshop.

Several methods for computing observer motion from monocular and stereo image sequences have been proposed. However, accurate positioning over long distances requires a higher level of robustness than previously achieved. This paper describes several mechanisms for improving robustness in the context of a maximum-likelihood stereo ego-motion method. We demonstrate that even a robust system will accumulate super-linear error in the distance traveled due to increasing orientation errors. However, when an absolute orientation sensor is incorporated, the growth is reduced to linear in the distance traveled, grows much more slowly in practice. Our experiments, including a trial with 210 stereo pairs, indicate that these techniques can achieve errors below 1% of the distance traveled. This method has been implemented to run on-board a prototype Mars rover.

We describe a method for recognizing surface-lying ordnance in test ranges using stereo range information and image edge maps. This method is to be used by an unmanned ground vehicle (UGV) surveying the test range for autonomous clearing of ordnance. We concentrate on a particular type of cylindrical ordnance (BLU-97) in current usage in U.S. military test ranges. In order to locate instances of the ordnance, we employ a stereo pair of cameras to be mounted on top of a UGV. Parallel segments corresponding to the occluding contours of the ordnance are detected in the imagery using robust and eecient model extraction techniques. The stereo range data is used to adaptively select the local scale for edge detection and to place constraints on the search space for the parallel segment extraction. Initial tests indicate that robust recognition is possible in near real-time with a low rate of false positives.

Least median of squares (LMS) regression is a robust method to fit equations to observed data (typically in a linear model). This paper describes an approximation algorithm for LMS regression. The algorithm generates a regression solution with median residual no more than twice the optimal median residual. Random sampling is used to provide a simple O(n log2 n) expected time algorithm in the two-dimensional case that is successful with high probability. This algorithm is also extended to arbitrary dimension d with O(nd − 1 log n) worst-case complexity for fixed d > 2.