tem-
> and
with
-Syn-
bing
per-
enti-
PLL
riva-
om a
pix-
eter-
id an
7CE
both
s for
an be
ernal
to be
, and
1 can
were
niza-
| pix-
first
ernal
naxi-
sev-
m-up
ther-
tions
ed to
nper-
:d by
sam-
1 ex-
es. It
erag-
e ac-
veral
limit
) use
Jlica-
"CE
e-di-
n the
' alu-
minium bars of a testfield structure. The targets span a
volume of 2.6 x 2.0 x 1.1 m (see Figure 12). The coordi-
nate system is placed such that the X and Y axes are
within the plane of the wall with the X-axis parallel to
the floor and the Z axis points away from the wall. The
coordinates of 86 targets were measured with theodolites
both before and after the imagery for the test was ac-
quired. The coordinates were determined with a preci-
sion of 0.02 mm in X and Y and 0.03 mm in Z. The
targets are photographically produced black circles with
a diameter of 20 mm on white background with a small
white dot in the center for theodolite measurements. The
centricity of the targets used for the theodolites is within
5 to 6 um of those used for the CCD-cameras.
Figure 11 shows the testfield with camera, theodolite,
Theodolite ?
Figure 11 Testfield, illumination, camera, and theodolite.
lights and curtains. The lights and curtains were placed to
generate a diffuse uniform illumination. The rods of the
structure and the camera fixture create visually almost
undetectable shadows on the wall. The local illumination
gradients are stable for shadows from the rods of the test-
field rods but vary their location for each position of the
camera stands.
The network and testfield are depicted in Figure 12. Im-
ages were taken with a SONY-XC77CE camera
equipped with a FUJINON 9 mm lens at 24 stations with
a roll of zero and 90 degrees. Both pixelsynchronous
frame grabbing and PLL line-synchronization was used,
each generating a set of 48 frames. This allows to investi-
gate the effect of the synchronization on the three-dimen-
sional accuracy. The image size is 728 x 568 pixel with a
spacing of 11 pum in x and y.
The image scale varies between 1 : 200 and 1 : 560. The
targets are thus imaged onto 3.3 to 9 pixels (diameter be-
tween inflection points of grayvalue profile) when ex-
cluding geometric foreshortening and other degrading
effects. The internal precision of the target location for
this size was experimentally determined to be in the
range of 0.005 to 0.01 pixel. The targets were imaged in
36 images on average.
Figure 12 Testfield and camera network.
Figure 13 Image grabbed at leftmost station showing the
large variation of image scale and the oblique-
ness.
3.2 Procedure and Mathematical Modelling
The pixel coordinates of all targets were measured semi-
automatically with least squares matching (LSM). The
pixel coordinates were subsequently transformed to pixel
coordinates with the pixel-to-image coordinate transfor-
mation (see Figure 14 for the definition of the coordinate
systems):
x = (x' -x) Xpsx (1)
y= (y,~¥) Xpsy (2)
with:
X, y ......-.-.-.....lrnage coordinates
X', y! .............pixel coordinates
x p! Y, Federer location of principal point in pixel
psx, psy .........pixel spacing in x and y
