04
ut
Its
ta
ell
rm
ics
ase
1ds
ect
red
the
iter
the
ted
1€
1Q
10
the
‚7m
ng
ned
Ake
ints
"ER
the
for
In a
a of
ext,
d by
1ean
are
the
etter
flat
>.
The
t, so
ging
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004
both scenes of a stereo model. During this time interval the
changes in the object space is often causing problems. The
SPOT 5 satellite includes with the HRS-sensor (High
Resolution Stereo) also a sensor which is generating a stereo
model within the orbit with 5m pixel size in the orbit direction.
These images are usually not available for standard applications
- SPOT IMAGE like to use these images only for the
generation of DEMs by themselves and like to sell only the
generated DEMs (see also Jacobsen 2004b), so usually only the
classical SPOT scene combination can be used for three
dimensional mapping.
In the area of Zonguldak a SPOT stereo combination taken at
August 13" and 14" of 2003 are available. Just one day of time
interval is optimal. Only below 1% of the area is covered by
clouds and cloud shadows and the atmospheric conditions are
the same.
E
E Y
Figure 4: histogram of SPOT 5 image Zonguldak
left: whole image right: forest
The panchromatic SPOT images do have also some problems
with the grey value variation in forest areas. Figure 4 shows on
the left hand side the histogram of the SPOT scene from August
14" with a mean value of 119 and a standard deviation of +/-
24.3 grey values. In a typical forest area the standard deviation
of the grey values is just +/-6.3. The second scene is very
similar.
shows the details of the distribution of the correlation
coefficients depending upon the areas. In the open areas the
correlation coefficient is quite higher like in the forests. The
darker areas — identical to a lower correlation coefficients — do
fit very well with the forest areas. Corresponding to this also a
lower accuracy of the generated DSM in the forest areas can be
expected.
Figure 6: quality map of image matching = SPOT 5-model
grey value 51 = correlation coefficient 0.6
grey value 255 = correlation coefficient 1.0
; =
; Vk |
SS
T TM * - ¢ #
+’ J
v
e
fü
oy
d AS ;
e
c =
pM v /
Eel 2
ii as
pon J T - e wm
- s 5 x Wr : N
za t a Cii he anat: i
Figure 5: matched points of SPOT 5-model Zonguldak
Matched points = white ^ Upper left: Black Sea
The scenes have been taken with incidence angles of 13.52? and
-16.65? corresponding to a height to base relation of 1.85. By
theory a height to base relation of 1.0 would be better, but such
a relation has stronger differences between the matching
windows of both scenes caused by objects above the terrain
surface like trees, buildings and rocks, so with the exception of
smooth areas, the generated DEM will have a similar accuracy
like models with a height to base relation of 1.0.
In figure 5 the accepted matching results with correlation
coefficients exceeding 0.6 are shown in white. The not accepted
points are located on the water and in forest areas. Figure 6
Level 1A (original images) and Level 1B images (projection to
a plane with constant height) of the same scene are available.
The bundle orientation of the level 1A images has been with
program BLASPO, while the different geometry of the level 1B
images required program CORIKON. The orientation results
are nearly identical, not indicating an advantage of one image
product against the other.
SPOT 5 DZ> | RMSZ | Bias RMSZ - bias
level 1A 50m [m] [m] F(slope)
All points 0.82% 13.5 7.1 93-57*tana
Open areas 0.87% 11.9 5.4 8.4+6.3*tana
Forest 0.76% 15.0 9.2 9.8 + 5.3 * tan a
Table 5: RMSZ of the SPOT 5 level 1A model against the
reference DEM from the map 1 : 25 000
SPOTS DZ» | RMSZ | Bias
level 1B 50m [m] [m]
All points 0.84% 132 72
Open areas 0.89% 11.6 5.4 8.6-4.3* tana
Forest 0.79% 14.9 9.4 10.0 + 3.3 * tana.
Table 6: RMSZ of the SPOT 5 level 1B model against the
reference DEM from the map 1 : 25 000
RMSZ - bias
F(slope)
9:5-- 3.9 mano.
The achieved root mean square differences of the DEMs based
on the SPOTS level.1A and level 1B models to the reference
DEM from the topographic map 1 : 25 000 are nearly identical.
No advantage of one image product against the other can be
seen. For the flat parts of the open areas the root mean square
Z-discrepancies corresponds to a x-parallax of 0.9 pixels.
639
