0.025 for scale. This two stage process produces a high
quality refinement. During the refinement, the time
necessary to compute the new correlation coefficient is
drastically lower than the time necessary to lag the target
through the entire search array. This is because the lag
position giving the best match is already known and the
resampled target array is correlated at this lag position
only. After the refinement procedure yields the minimum
coefficient, the x and y translation terms are added to the
whole pixel location to give the sub-pixel coordinates of
the panel. This procedure was also used for location of
fiducials.
The Mazomanie photography contained circular panels which
are insensitive to rotation thus eliminating the need for
rotation refinement. This reduces the time necessary to
initially locate the panel in the search array.
DIGITAL TERRAIN MODEL GENERATION
DTM generation is done by selecting target arrays on the
left photo in a regular grid pattern and correlating on
the conjugate imagery in the right photo. Knowing the
relative orientation parameters and approximate percent
overlap of the photos allows reasonably accurate prediction
of the search array on the right photo. In our test, for
comparison purposes, DTM points were selected as nearly as
possible to the location of points in the DTM read manually
in the PG-2.
When an acceptable match of the target array is found, the
photo coordinates of the conjugate images are computed.
Then the parameters of absolute orientation are applied to
compute the ground XYZ coordinates of the point. If an
acceptable match is not found, the target point is shifted
slightly and the process repeated. This is done twice and
if after that, the match is not found the point is rejected.
Once all of the points of the grid are correlated, the
coordinates are input to a contour mapping program and a
topographic map is plotted.
RESULTS
Using the LaCrosse photography an area of approximately 6
cm by 6 cm was selected in the left photo for DTM genera
tion. This area was in the middle of a field where there
were no trees present. The map resulting from this auto
matically generated DTM is shown in Figure 3. Figure 2
shows the map generated from the manually read DTM in the
same area. These two maps show reasonable consistency in
topographic features, however when they are overlaid, the
one obtained from the automatically generated DTM showed
elevations consistently higher than the stereoplotter map.
Possible causes for this shift are scanner distortions,
and inaccuracies of locating control points. The control
points in this set of test photography were located
manually by studying density overprints and were only
good to the nearest whole pixel. This problem should be
reduced with scanner calibration and control panel coor
dinate refinement used with the Mazomanie photography.