atching,
ilia. The
d image
nates by
d on the
o match
ference
f0.5-1
xels) in
10 km?
vely.
ing and
sensor,
erating
il 1993
ted and
ia with
from a
w of an
'S were
quality
at were
ng two
as the
f many
es, was
rse for
etween
caused
on the
raser et
e along
10-20
ia with
Figure l. 3D parallel view of the top left part of the fore channel (12 x 20 km). Orthoimage overlayed on automatically derived
DTM using a height exaggeration factor of 7.
The images, especially the nadir one, were of very poor
radiometric quality. The temperature on the Space Shuttle was
too high and this caused among other problems the application of
unsuitable sensor calibration parameters. In all images the
following problems existed: positive and negative spike noise,
pattern noise, and small grey value range (ca. 50 grey values were
occupied). The nadir channel had additional problems:
blemished lines (single or double, with very light or dark values)
in the left part of the image, different grey level mean in the left
and right part (the line CCD consists actually of two optically
butted CCDs which had different gain and offset due to the
aforementioned calibration problems), strong pattern noise at the
left of the left image part (different grey level mean every three
lines and every two columns), and higher grey level mean (ca. 25
grey values) than the mean of the fore and aft channels.
These radiometric problems are grave especially for DTM and
orthoimage generation. The control point measurement,
particularly if it is done manually, is not influenced so much.
Thus, for the control point measurement we did a strong contrast
enhancement and radiometric equalisation of the images using
Wallis filtering. For DTM and orthoimage generation the
following preprocessing was performed. For the fore/aft
channels: median filtering, generation of an image pyramid
(required for DTM matching), and Wallis filtering at each
pyramid level. For the nadir channel: grey level interpolation of
the blemished lines, median filtering, gain and offset
transformation of the right image part to fit to the left one,
smoothing of a narrow band along the border line of the left and
right part, image pyramid generation, and Wallis filtering at each
pyramid level. For the Oth pyramid level a Gaussian filtering was
111
applied before the Wallis filter to reduce the pattern noise and the
formation of grey level regions caused by the median filter. Some
results are shown in Figures 2 and 3 and speak for themselves. A
minor problem was due to the 3 x 3 median filtering: grey level
regions were formed in homogeneous areas (posterising), and
small objects (among them also some control points) were
distorted or partially eliminated. This could be reduced by using
a smaller support for the median filtering.
3. SENSOR MODEL
We used the model developed by Kratky, 1989. Kratky's model
processes single and stereo images of various sensors including
SPOT, Landsat, J-ERS 1 and MOMS-02 and is easily expandable
to include new sensors with similar geometries, as they become
available. It is an extended bundle formulation considering in a
rigorous way all physical aspects of satellite orbiting and of earth
imaging, together with geometric conditions of the time-
dependent intersection of corresponding imaging rays in the
model space. The ephemeris data (position and attitude) are not
necessary but for some sensors they may be used optionally.
Orbital perturbations are taken into account by allowing the
orbital segment to be shifted with respect to its expected nominal
position. The total number of unknowns per image is 14 - 6
elements of exterior orientation, linear and quadratic rates of
change for the rotation angles, a change Af for the camera
constant, and a quadratic distortion in x (corresponding to a shift
of the principal point along the CCD sensor). The quadratic rates
may be dropped (linear model). Six weighted constraints keep
the orbital positions of the two sensors within statistical limits
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996
