we
ding
ene
ding
123
the
| nO
> for
| an
S is
the
ible
rate
00
lere
the
the
jaw
ood
“to
Case 5: This is the most extreme case possible, using
only one control point in the first scene. As expected,
it is then not possible to estimate the orbit radius and
the yaw angle offset. This is most clearly seen in the
Y component where there is a trend in the across
track direction (Figure 7) corresponding to a missing
yaw estimate of 0.05 degrees. Still, the total rms
errors are kept close to 1 pixel.
5. ALTIMETRIC ACCURACY
5.1 Test area
A 25km x 25km area corresponding to a topographic
map sheet 1:50,000 was selected for evaluation of the
DEM computations. The area is situated in the
southernmost scene of the strip (scene 6). The terrain
is varied. Boreal forest with clearcuts dominates to the
east, low treeless mountains to the west, and a river,
bordered by agriculture, is running in the middle. The
height ranges from 300 m to 900 m. The image pair is
of good quality, except for a small cloud, and cloud
shadow, near the centre of the map area. The nadir
looking image over the map area is shown in Figure 8.
Figure 8. Nadir looking image (band 3), showing the
area covered by the reference DEM.
5.2 Test strategy
The JERS-1 stereo model was adjusted by the
method described in paragraph 3.1. The strategy of
the test was to set up different configurations of
control which were successively more difficult and
941
demanding. The numbering of the scenes is from
north to south, with number 6 being the one
containing the test map area. The following
configurations were tested:
1. Control points only in scene 6.
2. Control points only in scene 5.
3. Control points only in scene 4.
4. Control points only in scene 3.
4a. 4 control points only in scene 3.
4b. 3 control points only in scene 3.
4c. 2 control points only in scene 3.
4d. 1 control points only in scene 3.
The last test is most extreme, with only one control
point available 225 km away from the map area.
5.3 Altimetric results
The DEM's were computed using multi-point matching
(Rosenholm, 1987). The procedure is completely
automatic, and no post-editing was performed to
eliminate effects of the cloud. The DEM was
calculated with 200 m grid distance. The computed
DEM from test case 1 is shown as a grey-tone image
in Figure 9. The results were compared to the
reference DEM from the Swedish Land Survey.
Statistics in the form of mean and standard deviation
were calculated for the differences between computed
and reference DEM. The results are shown in Table
2.
Figure 9. DEM computed in test case 1.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996