Triangulation Based Hierarchical Image Matching for Mars Dem Generation Using Moc Na Stereo Images

Traditional image matching approaches often fail in processing Mars Global Surveyor stereo images because of low contrast and insufficient number of features. Reliable and precision matching tool is needed for precise Mars digital elevation model generation. This paper presents a hierarchical image matching approach. First, a number of well identified points are manually measured in a stereo pair. These measurements are input into a commercial digital photogrammetric tool as ‘seed points’ to generate more corresponding points. After that, the Delaunay triangulation network is formed for those seed points. The initial correspondence of an interest point on the other image is located by using the bilinear polynomial transformation whose coefficients are determined by six points closest to the interest point in the triangle. The conjugate point of the interest point in each triangle of the network is determined by using the parameters of the polynomial equation. In the next iteration, all the above generated corresponding points and the original seed points will be triangulated and the network will be densified in the same manner. This process repeats until the required point density is achieved. This initial prediction is then refined by using cross correlation and the least squares matching approaches. Based on this approach, we have matched stereo image pairs over selected landing sites for the Mars Exploration Rover mission. Assessment of image results suggests a matching consistency of better than one pixel. Presented in this paper are results of detailed matching quality evaluation and the generated digital elevation models.