RELATIVE ORIENTATION
Relative orientation is necessary to derive DTM coordinated
in model space. Pass points are selected on the left photo
in the traditional locations and correlated to conjugate
imagery in the right photo. An analytic solution is then
performed giving the relative orientation of the two photos.
This facilitates location of the DTM points by enabling use
of epipolar geometry which reduces the size of the search
array containing the matched point. The principle of epi
polar geometry states that conjugate imagery must lie along
the line of intersection between the photo plane and the
plane defined by the perspective centers and model point.
Exploiting this principle enables a much smaller search
array to be used thus reducing the computational time
necessary to locate conjugate points.
AUTOMATIC CONTROL PANEL RECOGNITION
To enable the derived DTM coordinates to be transformed to
the ground system, control points must be located and their
photo coordinates determined. For a panel to be automati
cally located, an array must first be synthesized which
"looks like" the panel on the digitized photo. Since the
panels can be in any orientation relative to the scan
lines a single standard array is not sufficient and a means
of simulating different scan line orientations is necessary.
The method of resampling known as bilinear interpolation is
used in conjunction with a two-dimensional conformal
coordinate transformation to simulate the different
orientations. A standard panel array, with the wings of
the panel exactly coincident with scan lines is used as a
basis to generate many possible orientations until one is
found which is close to the actual orientation of the panel
on the scanned image. The algorithm for finding control
panels is as follows. First the standard panel array is
lagged through the search array. If the minimum correla
tion coefficient is unacceptable the standard panel array
is rotated 45 degrees by means of the two-dimensional
conformal coordinate transformation and the pixels are
resampled. This new array is lagged through the search
array and if the coefficient is still unacceptable further
rotations are performed until an acceptable match is found.
If after a specified number of trial rotations the panel
has not been found the point is rejected. Once the coef
ficient is within the threshold a fine tuning procedure
is used to refine the location of the panel to sub-pixel
coordinates. In the fine tuning process, parameters of the
two-dimensional transformation are incremented or decrement
ed one at a time by small amounts, the panel array is
resampled and a new correlation coefficient computed.
After a minimum coefficient is obtained by a series of
incremental shifts of one parameter, the next parameter is
incremented and decremented until a new minimum is reached.
The order of the parameters refined is first rotation by
two degree increments, then x translation, y translation
and scale all in increments of 0.1. A contrast stretching/
reducing term is then varied in the same fashion and the
whole process starts over with smaller increments of 0.5
degrees for rotation, 0.025 for x and y translations, and