points were really necessary, provided a best gravity model for 
the planet is available to model the orbits. 
2 DEM Image Match Poin 
Image matches are obtained by normalized cross-correlation of 
overlapping images. While radar images may call for radar- 
specific radiometric and geometric features in an image 
matching system, such algorithms do not exist. Instead, regular 
cross-correlation matching is being used in initial work (Leberl 
et al., submitted). 
3.3 Intersecting Terrain Points 
A pair of image match points (x y x " y ) results in a surface 
point XYZ based on a sensor model that converts the image 
measurements into a slant range and Doppler cone, and 
associates this with the spacecraft's state and velocity vector. 
3.4 Ortho-Rectifving the Images 
The xy and XY(Z) coordinates are the basis for ortho- 
rectifying the images into a planetary map coordinate system, 
so that an image mosaic emerges from combining individual 
ortho-rectified images into image map quadrangles. 
3.5 Gridding the DEM 
The computed mass points at their XYZ locations need to be 
converted into a regular grid of DEM points by an interpolation 
process. 
litv A ment and Editin 
Of course any extracted data need to be verified for their 
accuracy. There are several options: 
(a) Visual inspection and manual editing; 
  
(b) Figures of merit from; 
image matching; 
stereo-intersection. 
(c) Use of overlaps in stereo image pairs and 
assessment of internal consistency. 
Refining the DEM with Shape-From-Shadin 
The topographic expression inherent in radar images has been 
the reason that geological users of radar-stereo DEMs have 
been able to manually correct DEMs: the geological expert 
will use geomorphological details visible to the human 
interpreter of the images, but this detail is not available from 
stereo-parallaxes. 
This has been the driving force behind the development of a 
DEM-refinement scheme based on converting local brightness 
differences to local slope differences. This has shown to 
improve the local morphology of the DEM (Thomas et al., 
1991). The process represents an automated DEM-editor or 
“air brush” to add detail on the pixel-level to a DEM that 
otherwise would have postings every 10 to 15 pixels. 
3.8 Going From Image to DEM 
The process described in subsections 3.1 to 3.7 avoids the use 
of a real-time math-model for an analytical plotter. Instead the 
DEM is produced as if a stereo comparator had been employed. 
The on-line real-time relationship between the planetary XYZ 
; F7 "on . . 
system and image spaces x,y ,x ,y could be desireable in an 
interaction with the stereo images, but is not needed in the 
creation of DEMs and image maps. 
4. SAMPLE RESULTS 
Figure 3 is the perspective view of a DEM created from the 
stereo pair in Figures 2a, b. At the time of this writing, an end- 
Figure 3: Perspective view of a stereo-derived DEM of 
2000 x 6000 pixels covering 150 x 150 sq. km. 
Area covered by Figure 2. The image is draped 
over the DEM. Elevation differences are in the 
range of 2 km, and slopes of more than 30° 
occur. 
mh AS oud heed (QUU yg AY A ep 
bt d o m o bud 
pede gm IN PP ie ep | 7T