Figure 8. GCP’s and Tp’s Used in this Work 
Table 8 is a subset, at the end of the iteration sequence, from the “view of triangulation results”, while Table 9 
from the “view error propagation reports”, respectively. Only 17 GCP’s, of the 27 available at that moment, with 
all 26 tp’s were selected in the final triangulation process -see Fig. 8. 
From the triangulated images, the Stereo Tool, using a third degree “warping polynomial”, produces a 
stereo pair or epipolar images. At this level a Reduced Resolution Data Set (RRDS) is generated averaging every 
four pixel (2x2) down to a single pixel and iterating until an RRDS is generated which is less than 64x64 pixel; 
this new imagery is used as an overview for image selection, broad area coverage display and automatic DEM 
collection. The DEM Tool, using RRDS’s to reduce false fixing, digitally correlates the orthorectified patches of 
imagery, on the base of a hierarchical approach, to produce the DEM of the overlapping area. The results are 
dependent on imagery scale (the higher resolution the more difficults), image quality and terrain roughness (the 
more rugged terrain or dense cultural details the lower correlation success). At the end of the process a log-file is 
produced. Snow covers and bodies of water produces bad collection of points to be used in the DEM production: 
in our case, e.g., on the water was collected a point at -1814m deep, That was simply corrected by hand. At the end 
the Orth Tool is used to produce ortho images. 
In an effort to verify the accuracy of the DEM produced, a comparison was made with a DEM generated 
from contour lines of the same area, vectorized into an ARC/INFO coverage. To do that, a “triangulate network 
and, from that, a “lattice" were generated with ARC/INFO facilities. Using Imagine (ERDAS) one DEM was 
subtracted from the other. A test area was selected over which statistics were made - see Table 10 and Fig.9. The 
gray scale difference was then converted numerically and superimposed on the right image (Fig. 10). Considering 
the other image (Fig. 11) it can be seen that the higher errors (the maximum is 289m) are not in white areas but in 
the upper left part due to the lost of details and to the difference in radiometric quality of the second image. 
185