tanbul 2004
ofiles of the
cture of the
) shows the
M and the
file is very
)EM (Fig. 5
| reference
£M WITH
A
e reference
iuttle Radar
vailable for
vs the mean
1x values of
Barcelona.
I-/SRTM-
Min / Max
-47 | +37
-59 / +53
-62 / +63
158/+191
158 / +191
-22 / +25
-98 / +62
218 / +135
484 | +394
484 / +394
ases, while
with more
, with very
ion and the
a coarse
fference in
due to less
RTM-DEM
s due to the
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B1. Istanbul 2004
sea level calibration. But it can be seen clearly that the standard
deviations and min/max values become significantly higher in
mountainous regions, which could be expected due to various
effects (shadowing, foreshortening, layover etc.) and more
difficult matching situation. In flat or moderate terrain both
models are of similar quality and can easily be used for DEM-
fusion.
A DEM fusion has been performed with the support of accuracy
layers from both DEM data sets. In the SRTM case this layer is
produced on a routine base by using features of coherence and
density of residuals in the DEM generation process. For the
optical data, an accuracy layer was generated by using the mean
standard deviation as a lower limit and the density of the
matched points after the region growing process. The fused
DEM shows lower standard deviations especially in moderate
and mountainous terrain. Table 5 shows the results of the
comparison of the reference DEM to the fused DEM. The
absolute height was taken from the SRTM-DEM, therefore the
mean height difference is as low. The standard deviation is
improved in all cases compared to the DEM produced only by
one method. This result shows that the usage of several DEM
from different sources of similar quality can improve the overall
quality.
Table 5: Area-wise comparison of height of FUSED-DEM
and reference DEM
Reference . | Size of Mean Height | STDV | Min / Max
area Area Difference [m] | [m]
Barcelona City | 71 km? 0.9 3.7 -23 / +28
Rural Area 161 km? -1.3 49 | -62/ +48
Moderate 105 km? -1.0 56 | -70/+78
Mountain
Montserrat 84 km? -1.5 11.1 |-183/ +201
Whole area 623 km? -1.0 5.7 1-183 / +201
8. BUNDLE BLOCK ADJUSTMENT
From the results of the analysis based on the orientation data
provided in ancillary data files it can be seen that small biases
in all three coordinates, in the order of an HRS pixel size, still
remain. These can be removed with the help of a few ground
control points using bundle block adjustment. For the bundle
block adjustment the software CLIC, developed by TU Munich
is used (Kornus 1997). To apply the CLIC interior orientation
model the values of the model parameters have to be derived
from the given look angles of CCD elements. The type and
values for the CLIC interior orientation parameters are given in
table 9. Focal length and pixel size are taken from Gleyzes et al.
2003). Standard deviations are estimated. In the MOMS case a
CCD curvature parameter was additionally used, which was not
introduced for SPOT because an equal distribution of GCP over
the whole image swath is necessary for its determination which
was neither available for the Bavarian nor for the Catalonian
test site.
Table 6: Interior orientation of HRS1/2 as input to bundle
adjustment
HRS] o HRS2 G
f focal length [mm] |580.5 0.01. 1580.3 0.01
Xo |principal point |0 0.5 0 0.5
yo joffset[pixel] 6(39 um) 0.5 16 (104 pm) [0.5
9 rotation of CCD [°] |-0.05649 0.001 | 0.00735 0.001
ly stereo angle [?] |20.0378 0.001 |-19.95754 0.001
Because of the special optics employed for HRS the parameters
of table 6 cannot fully describe the interior orientation. The
remaining deviations are shown for HRSI in figure 6. These
values are introduced into CLIC as ‘synthetic’ calibration
tables.
All calculations are done in the LTS. Exterior orientation for
HRS1 and HRS2 is directly used as specified by the image
providers with the following standard deviations:
0,7 06,70, -0.5m and 0,7 o, = o, - 0.000005?
The relative accuracy of all exterior orientation values is
therefore taken to be very high. More variation is allowed for
some bias parameters of the exterior orientation which can be
modelled separately in CLIC. Because of the nearly constant
offset seen in the shifts between the orthoimages only the bias
of the pitch angle ¢ for sensor HRS2 is specified with a higher
standard deviation of 0.002° (input value for the bias itself: 0°).
This value has been estimated from the shifts of the
orthoimages. The bias values for x, y, and z have been set to
0m with a standard deviation of 1 m. For each camera 15
orientation images (857 image lines for one orientation interval)
have been used for modelling the exterior orientation.
Figure 6. Residuals of the HRS1 camera model fed into
CLIC as “calibration file”
From mass tie points generated by DLR matching software a
well distributed subset of about 15000 points is introduced into
CLIC. In the Catalonian case 28 ground control points are
measured in six of the orthoimages provided by ICC. First, the
GCP image coordinates are measured manually in the HRS1/2
images, and then the HRS2 coordinate is refined by local least
squares matching to subpixel accuracy. For the Bavarian case
10 GCP are used. All GCP map coordinates are introduced into
CLIC with standard deviations
6,7 0, 10m. add, o, 5m
Standard deviation of image coordinates (tie points and GCP):
6, = o, = 0.2 pixel for tie points and
6, = 0, =0.3 pixel for GCP
Table 7 gives the values of the interior orientation parameters
after bundle adjustment. No substantial changes can be
detected.
Table 7: Interior orientation after adjustment
Bavaria Catalonia
HRS1 HRS2 HRSI HRS2
f (mm) | 580.50407 | 580.29826 | 580.49962 | 580.29999
Xo (um) -0.465 -0.528 0.642 0.250
yo (um) | 40.760 89.790 37.426 103.473
à (deg) |-0.05130° 0.01196° |-0.05606° 0.00734°
Y (deg) | 20.03824° 1-19.9570° 120.03718° (-19.9577°
