ilt
lage
to
of
gle
111
for
of
The
of
is
be
ust
age
ion
ger
ata
ect
nce
and
CT,
ing
are
orm
age
ric
ime
ow.
of
me)
se.
es,
for
and
ted
red
ge,
nto
his
ely
of
white "blobs" on black background, or vice versa. The feature extraction
process isolates each blob and gives it a unique label. This operation is
carried out by library functions applying a connectivity-analysis pixel
grouping algorithm.
In order to distinguish the blobs representing the targets from
other blobs, characteristic parameters are computed for each one. The
differences between these computed values and a given set of parameters
for the ideal target are computed and the blob is recognized as a target
if the differences are within a pre-set tolerance.
Each recognized target covers an area of several pixels and thus
it is necessary to locate, with sub-pixel accuracy, the coordinates of the
center of the target. For solid-colored symmetric targets of highly
contrasting background, the coordinates of the centroid are computed and
considered to be the target coordinates. However, if the targets are
designed with their center points clearly marked, a second step is carried
out. In this step, the enhanced image is recalled and the grey levels of
the center pixel and a matrix of pixels around it are used to interpolate
the coordinates of the center.
All the previous computations are applied to individual images,
one from each camera. Now each point in one image must be matched with its
corresponding point in the image taken by the other camera. For the
control points, which are needed to determine camera orientation and
calibration parameters, a priori knowledge about the number of these
points and the way they are arranged is required. For all other points, to
match a point in an image with its corresponding point in the other, the
image coordinates of this point in the first image and the orientation
parameters of the two images are used to determine the relationship
between the x- and y-coordinates of this point in the second image (figure
VIII-A). This is a straight line relationship (the epipolar line) as shown
in figure VIII-B. The image coordinates of all the targets in the second
image are tested with the equation of the epipolar line and the target
that satisfies the equation best is the best match. However, this best fit
must be within a preset tolerance, otherwise the point has no match and
will be eliminated.
Once all the targets are matched, their image coordinates and the
orientation parameters are used to determine the object coordinates of
these targets by photogrammetric intersection.
SUMMARY
In conclusion, we have provided a brief description of the digital
image processing facilities currently available at NRCC/PR and have
discussed the application of one of the systems to a fully automated
photogrammetric task. While none of the equipment is of exceptionally high
performance calibre, we feel that modest equipment such as ours is capable
of addressing a wide range of problems relating to photogrammetric digital
image processing. It seems certain, on the other hand, that product
oriented system development for an all digital photogrammetric system will
require equipment pushing state-of-the-art technology if implemented in a
straightforward. manner. Future research may illuminate techniques for
lessening the technological requirements or, at least, provide effective
and efficient digital solutions to photogrammetric problems.
159