y data
oss of
naster
e gray
e dark
ausing
2 blue
CCD-
for all
juest for
ry 1995.
it source
ed. with
le. The
arts are
cing the
M1
The general geometric design is corresponding to an
analytical plotter. The movement of the photo carrier is
controlled by servos based on linear encoders. The plate
is moved by a friction drive. With a working space of
245mme245mm no problems for handling standard size
aerial images are existing.
The used time delay and integration sensor (TDI) DALSA
CL-E1-2048A has a length of 2048 elements and
integrates over 96 pixels, that means, it is not a line, it is
an array with 2048e96 elements. The integration of the
TDI-sensor over 96 elements reduces the noise and
systematic line errors by the factor of 96 and raises the
sensitivity 80 times against an individual CCD-element.
This sensor is also used in another photogrammetric
image scanner.
The RM1 is using only 1024 of the 2048 rows. The active
part can be selected by the scanner control data, usually
the center part is used. The sensor has no possibility for
a selection of the spectral range, this only can be done by
a computer controlled filter wheel with the standard
red/green/blue selection. The transmitted light from the
fluorescent lamp is passing through an APO-RADOGON-
D lens system from Rodenstock over a mirror to the
sensor. By the projection the original pixel size of 13yum in
the sensor is changed to 12um in the scanned image.
3. GEOMETRIC ACCURACY
If the digital images shall not be used just for orthophoto
production, the loss of accuracy caused by the scanning
is important. The geometric quality of the scanner should
correspond to analytical plotters, it has the same meaning
to the data acquisition. The geometric accuracy should
not be mixed with the pixel size because a subpixel
standard deviation can be reached. If the image target
has a size of one pixel, by theory the location is defined
with +0.3 pixel, but usually the elements are defined by a
higher number of pixels. Well defined, targeted points can
be measured with up to 40.02 pixels (Bósemann,
Jacobsen 1995).
The basic construction of the RM1 is corresponding to an
analytical plotter, but there are few additional error
sources because opposite to the analytical plotter the
image coordinates are not just determined for a point, a
line of pixels is scanned at the same time. The lack of
flatness of the photo on the photo carrier has only an
influence corresponding to the view direction of the
sensor line. the center part of the sensor is used, in
maximum the inclination of the imaging ray is
corresponding to 6.144mm / 150mm = 1:24. That means,
a lack of flatness of the photo of 0.1mm is causing a 4 um
dislocation in maximum, in the linear mean 2pm.
The geometric effect of the sensor line itself is negligible,
it is far below 1um. The measuring system is based on
servos together with a glass scale from RSF Elektronik. A
resolution of 0.5um is reached and the used type
MSA6707 shall have an accuracy of +/-3um. The x- and
the y-axis should be orthogonal, this has to be calibrated.
Finally we do have the same problem like in analytical
plotters. Usually an error in affinity and angular affinity is
unimportant because it is eliminated by the inner
orientation which has the be done at least by affinity
transformation in relation to the fiducial marks because of
affine errors of the original photos.
The location of the sensor line in the image has to be
calibrated. The swath width shall correspond to the
distance between the neighbored scan lines and the
sensor line shall be orthogonal to the scan direction (see
figure 2).
The RM1 has been checked by means of 2 different
reseau platen, one from the Rollei Reseau Scanner with
121e121 reseau crosses with a spacing of 2mm and one
with 11e11 lines with a grid spacing of 12.5mm. For both
platen the calibrated grid coordinates are available with
an accuracy < + 1um. The platen have been rotated
slightly against the instrument axis to avoid the check of
only few rows and columns. At first the reseau cross
positions in the scanned images have been measured
manually with a pointing error of +2um, later on this has
been done by correlation with an accuracy «1pm.
The discrepancy between the measured and transformed
grid coordinates to the calibrated values have been
analyzed for random and systematic errors. The
separation of the systematic errors can be made in the
same way like the self calibration in bundle block
adjustment. Additional unknowns which are able to
compensate the typical geometric problems have to be
introduced.
<>
*— | swath width
misclosure in y-direction
[e
—
= sensor line
|
v
p gap between
neighbored scans
Figure 2: location of CCD-
lines of neighbored scans,
errors of mechanical sensor
calibration
sensor line not
orthogonal
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B1. Vienna 1996
