HR: 0830h
AN: P41B-0411    [PDF]
TI: Automated Extraction of Planetary Digital Elevation Models
AU: * Andre, S L
EM: sarah@earth.northwestern.edu
AF: Northwestern University, Department of Geological Sciences, 1850 Campus Drive, Evanston, IL 60208 United States
AU: Andre, T C
EM: troy@hep.uchicago.edu
AF: University of Chicago, Department of Physics, 5640 S. Ellis Avenue, Chicago, IL 60637 United States
AU: Robinson, M S
EM: robinson@earth.northwestern.edu
AF: Northwestern University, Center for Planetary Sciences, 1850 Campus Drive, Evanston, IL 60208 United States
AB: Digital elevation models (DEMs) are invaluable products for planetary terrain interpretation [i.e. 1,2,3]. Typically, stereo matching programs require a user-selected set of corresponding points in the left and right images (seed points) to initiate automated stereo matching routines, which then find matching points between the two images. User input of seed points for each stereo pair can be a tedious and time-consuming step. An automated stereo matching tool for planetary images is useful in reducing or eliminating the need for human interaction (and potential error) in choosing initial seed points. In our software, we implement an adaptive least squares (ALS) correlation algorithm [4] and a sheet-growing algorithm [5]. The ALS algorithm matches a patch in the left image to a patch in the right image; this algorithm iteratively minimizes the sum of the squares between the patches to determine optimal transformation parameters. Successful matches are then used to predict matches for the locations of surrounding unmatched points (sheet growing algorithm). Matching is initiated using either automatically generated seed points or manually picked seed points. We are developing strategies to identify and reduce the number of errors produced by the stereo matching software; additional constraints may be applied after the matching process to check the validity of each match. We are currently testing the stereo matcher on image pairs using correlation patch sizes ranging from $9\times 9$ pixels to $25\times 25$ pixels. A rigorous error analysis will be performed to better assess the quality of the results. Initial results of DEMs derived from Mariner 10 images compare well with DEMs generated by another area-based stereo matcher [6]. Our ultimate goal is to produce a user-friendly, robust stereo matcher tool that can be used by the planetary science community across a wide variety of image datasets. [1] Herrick R. and Sharpton V. 2000, JGR 105, 20245-20262. [2] Oberst J. et al. 1997, Eos 78, 445-450. [3] Smith D. et al. 1999, Science 284, 1495-1503. [4] Gruen A. 1985, S. Afr. J. of Photogramm. Rem. Sens. Cart. 14(3), 175-187. [5] Otto G. and Chau T. 1989, Image Vision Comput. 7, 83-94. [6] Cook A. and Robinson M. 2000, JGR 105,9429-9443.
DE: 1224 Photogrammetry
DE: 5400 PLANETOLOGY: SOLID SURFACE PLANETS
DE: 5464 Remote sensing
DE: 5494 Instruments and techniques
SC: Planetary Sciences [P]
MN: 2003 Fall Meeting