a rov-
> been
19..For -
Land-
allel to
D pro-
ntation
er look
to the
e lines
for all
=
1200
le from
Each line has creases at approximately row 650 and
row 1000 (at a pixel size of 13.5 m). This is a con-
sequence of the surveying process in which the mea-
surement of the whole profile was subdivided in three
parts with two control points in between. Obviously,
in this profile the horizontal coordinates between ad-
jacent control points have been calculated with a 5 m
spacing interval and the appropriate height data are
determined by the GPS measurements.
An impression of the DTM quality is given by figure
10. Along the profile the DTM height is interpolated
and the difference to the GPS profile is calculated. A
mean difference of about 60 m is subtracted which in-
dicates that there are still some systematic influences
(see discussion below). The remaining differences
(figure 10) are in an interval of + 10 m with an accu-
racy of 7.6 m.
-—
a
>
e
SH
de ENE!
a
height differences: GPS - DHM [m]
Ty EE Ji
0 2000 4000 6000 8000 10000 12000
profile [m]
Figure 10: Height difference between GPS and DTM
measurement along the profile
A further improvement of the DTM quality is achieved
with the developed area based least squares solu-
tion for MOMS three-line images. The basic con-
cept of this approach is to match the three channels
in various combinations which provides high quality
measurements and allows a sensitive selfdiagnosis.
For more details please cf. Fritsch et al. (1995) and
Schneider and Hahn (1995).
The matching result with this area based solution
along the DTM evaluation profile is shown in figure
11. Displayed are the differences between the pro-
jected GPS coordinates and the measured image co-
ordinates for the backward and the forward looking
channels. The projected point location of the 3228
DTM check points in the nadir channel is taken as
Seed points to start the matching.
271
e e zm
oO e e
1
—
a
Diff in (x, y): GPS - Matched [Pixel]
N
e
0 2500 5000 7500 10000 12500 15000
profile [m]
Figure 11: Coordinate differences between GPS and
the matching result
So far we still observe some inconsistency with the
orientation and calibration data which shows up in
a systematic displacement of up to 2 pixels in im-
age space. Because the reason for this displace-
ment could not yet fully be explored the uncor-
rected differences between the GPS points and the
matched points are plotted in this figure. Possible er-
ror sources are in an incorrect use of the given ori-
entation data and in the definition of the datum. It is
important to note, that the systematic displacement is
constant and therefore can be eliminated. Substract-
ing an arithmetic mean value, the following standard
deviations are found: 0.23 pixels in x and 0.21 pixels
in y direction of the forward looking channel and 0.13
pixels in x, 0.15 pixels in y in the backward looking
channel. Altogether the selfdiagnosis process of the
matching module indicated that 92.7 96 of the 3228
profile check points have been matched successfully.
In a small region of about 150 GPS points (55 pix-
els, visible in figure 11 at profile length 12500 m) the
matching failed. The reason for this failure is a blem-
ished line which can be seen in figure 12.
In general a preprocessing should eliminate this kind
of line distortions (which occurs only in the nadir
channel). Unfortunately this was not carried out thor-
oughly in this area. On the other hand this example
shows that the selfdiagnosis of matching procedure
works well thus erroneous matches caused by this
distortion are avoided.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996
