期刊名称：ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
印刷版ISSN：2194-9042
电子版ISSN：2194-9050
出版年度：2002
卷号：XXXIV Part 4
出版社：Copernicus Publications
摘要：High-precision topographic information from all available data is crucial to many landing site geological and engineering applications. At the same time, precise navigation and localization of the rover as it traverses the Martian surface is important both for its safety and for the achievement of its engineering and scientific objectives. In this paper, we investigate a new approach to high-precision Mars landing site mapping and rover localization based on integrated bundle adjustment of an image network built by linking ground-based images with automatically or manually selected tie points. The method and software are tested using lander and rover image data obtained from the Mars Pathfinder mission. An innovative method for automatic tie point selection is also presented. 1. INTRODUCTION High-precision global and local topographic information is crucial for support of engineering operations and achievement of scientific goals in Mars exploration missions. In particular, landing site mapping will be extremely important for landing and rover navigation in future lander and rover missions such as the 2003 Mars Exploration Rovers (MER), the European Beagle 2 lander of the 2003 Mars Express, the 2007 Smart Lander and Long-range Rover, and sample return lander missions beyond 2010. In Mars rover missions, accurate navigation and localization of the rover relative to the landing center are needed so that the rover can traverse the Martian surface safely, return repeatedly to the same location to perform operations, and support coordinated multidisciplinary high-precision scientific experiments. In the Mars P athfinder (MPF) mission, the rover Sojourner principally used a heading sensor and wheel counters for localization. Sojourner accumulated 1m location errors in an area of about 10m x 10m during this mission. In the MER 2003 mission, rovers will traverse an extended distance of 600m up to 1,000m where the terrain may be more challenging. In the 2007 Smart Lander and Long-range Rover mission, the rover traverse will be even longer. More accurate rover localization will be highly desirable for achievement of the scientific objectives of future missions. In order to prevent the accumulation of localization errors, descent and orbital images should be used as well to provide global constraints. Since 1998, the Mapping and GIS Laboratory at The Ohio State University and the NASA JPL Machine Vision Group have been jointly developing a bundle adjustment method with relevant techniques for the processing of Mars descent and rover images for rover localization and landing site mapping. We missed an opportunity to test the concept using actual MARDI (Mars Descent Imager) data because of the failure of the Mars Polar Lander Mission. In order to verify our algorithm and software, field tests were conducted at Silver Lake, CA in April 1999 and May 2000. Based on the data obtained, various rover localization experiments were carried out. Using descent and rover images along with either an integrated or an incremental adjustment, rover localization accuracy was reached of approximately 1m for a traverse length of 1km from the landing center (Li et al., 2000, 2001; Ma et al., 2001). Experiment results also showed that it is feasible to localize the rover by using rover images only if no descent images are available (as with the MER mission), and yet still achieve a similar accuracy when issues such as optimal traverse design and image network generation are considered (Di et al., 2002). In addition to using simulated descent and rover images, we tested our methods and software with actual Mars data. We downloaded rover and lander IMP (Imager for Mars Pathfinder) images from the PDS (Planetary Data System) web site. The German Aerospace Center (DLR) also provided us with a complete panorama chosen from a vast mount of IMP images. In the following section we briefly describe the bundle adjustment models used in this investigation and our other relevant studies. In subsequent sections, the processing of IMP panoramic images and rover images are presented. Symposium on Geospatial Theory, Processing and Applications, Symposium sur la théorie, les traitements et les applications des données Géospatiales, Ottawa 2002 var currentpos,timer; function initialize() { timer=setInterval("scrollwindow()",10);} function sc(){clearInterval(timer); }function scrollwindow() { currentpos=document.body.scrollTop; window.scroll(0,++currentpos); if (currentpos != document.body.scrollTop) sc();} document.onmousedown=scdocument.ondblclick=initialize 2. BUNDLE ADJUSTMENT MODELS The basic model for the bundle adjustment is based on the well known collinearity equations (Wolf and Dewitt, 2000). The linearized observation equation is expressed in matrix form as: L - AX V = (1) In this bundle adjustment model, all of the unknowns (camera position, orientation of all the images, and 3D ground position of the tie points) are adjusted together after all the images are acquired. Therefore we call it the integrated bundle adjustment model. Because there is no absolute ground control available on the Martian surface, the adjustment is a free net adjustment where the normal matrix is rank deficient. We used the Singular Value Decomposition technique to solve the normal equation in which the Minimum Norm is applied using the Least Squares principle. If certain distinctive features (such as mountain peaks or craters) can be observed from the orbital images, they are used as relative controls for the adjustment in the following form: W HX = (2) In order to improve computational efficiency and make the bundle adjustment applicable to a real-time operation, we developed two incremental bundle adjustment models (Li et al., 2000, 2001; Ma et al., 2001). Model I is represented as: m m m m m m 1 - m 1 - m 1 - m 1 - m L - Y B X A V , L - X A V + = = (3) Model II is represented as: m m m m m m 1 - m 1 - m 1 - m 1 - m 1 - m 1 - m L - Z C Y B V , L - Y B X A V + = + = (4) In Model I, the unknowns are added gradually into the adjustment system when the new observations are available. After the step m adjustment, unknowns solved by step m-1 are also updated so that the final results are ultimately the same as those from an integrated bundle adjustment model. In Model II, with the gradual increase of new unknowns some older unknowns are no longer taken into account, thus offering more flexibility and efficiency to the process. There are several frames of reference that are used in the MP F image pointing data, including the camera head coordinate system, lander (L) frame, Martian Local Level (M) coordinate frame, Mars Surface Fixed (MFX) coordinate frame, and Landing Site Cartographic (LSC) coordinate system. Bundle adjustment and topographic products of our models are based on the LSC system as defined by the U.S. Geological Survey. We developed a program to convert the pointing data of the PDS images to the exterior orientation data. This is accomplished by a chain of translations and rotations through the above frames of reference. Converted exterior orientation values were then used as initial values in the bundle adjustment. 3. PROCESSING OF IMP IMAGES We used 60 pairs of IMP stereo images provided by DLR to explore automatic tie point selection, bundle adjustment, DEM (Digital Elevation Model) and orthoimage generation. These IMP stereo images form a complete panorama (as shown in Figure 1), in which the azimuth and the tilt angles of the images are represented in the x and y axes, respectively. Figure 1. Coverage of the IMP stereo images Figure 1. IMP image coverage We developed a series of techniques to automatically select tie points within one stereo pair (intra-stereo) and between adjacent stereo pairs (inter-stereo). For intra-stereo tie point selection, the F.rstner operator is applied to extract interesting points. Then, an area-based matching is applied to match these interesting points using normalized correlation coefficients. Experiments with this data set show that in most cases over 90% of the matches can be found. Figure 2 shows the matched (white cross) and unmatched (black cross) interesting points in an intra-stereo pair. In this stereo pair, 319 and 312 interesting points were extracted from the left and right images, respectively, and 216 points were matched. Figure 2. Interesting point matching Within a set of matched points, some mismatched points may still exist that need further verification. Automatic verification of the matched tie points is based on the consistency of parallax. Assuming that the terrain does not vary significantly, parallaxes of the points should maintain an order throughout the image. In other works, if two points in the left image are located at the top left and bottom right corners, respectively, then their matches should preserve the same order (e.g. measured by a distance). Disordered points are considered as mismatches. The order can be expressed simply as a function of parallaxes along the row direction, since in the column direction the parallaxes are almost a constant (close to zero). In Figure 3, we order the interesting points in the row direction and then draw their x (left) and y (right) parallax curves in the same order. Any abrupt changes of the parallaxes are assumed to represent mismatches. Mismatches are eliminated using a median filter to produce the smoother parallax curves found in Figure 4. These smooth parallax curves represent the correct order, and are used as reference parallaxes. Figure 3. The x and y parallax curves
关键词：Mars Pathfinder Mission; Landing Site Mapping; Rover Localization; Bundle Adjustment
