A Case Study in Extracting DEMs from High-Resolution Mars Stereo Pairs Using a Simple Computer Vision Algorithm

Background: Geoscientists have been doing research on Mars for decades. As the photography technique has improved, we now have thousands of high-resolution stereo pair images but it is difficult to extract quantitative information including relative heights and slopes without the aid of computer vision. Our goal is to create an easy-to-use, accurate and high performance tool which can process a medium size(6000x6000) Mars stereo pair on a standard desktop, and transforms stereo pairs into digital elevation maps for geoscientists. Method: We applied a simple computer vision algorithm which could be implemented in C++ in 1500 lines to compute elevation maps from stereo pairs. The computer vision algorithm produced a disparity map and we used position data from the satellite that produced the stereo pair to create a digital elevation map with height measures in meters. Conclusion: We ran three tests, Victoria Crater, Centauri, and Gale Crater stereo pairs. The details of the stereo pairs are listed below. The accuracy values are the comparisons between our results and the ground-truth data provided by NASA[3]. Name Detail Victoria Crater Centauri Gale Crater Image Size 2000x2000 6824x6236 14011x13099 Maximum Height Difference (computed with SOCET Set software) 79m 477m 506m Computational Time Using Our Tool 45min 22hours ~5days Our Algorithm’s Accuracy compared with SOCET Set result (measured by differences less than 15 meters) 94.97% 89.11% 87.46% Figure 1. Our challenge was to make it easier to convert a stereo pair (left two images) into an accurate digital elevation map presents as a image. (rightmost image). Author et al: MARS 1, 1-6, 2005 2 of 8