2004
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B1. Istanbul 2004
"preventive regularisation". The term regularisation comes
from the definition of a "regular matrix", i.e. a full-rank matrix,
i.o.w. an invertible matrix. Alike, a singular normal equation
matrix has to be made regular before a solution may be
obtained. Such singularities may occur in our context when:
— Not enough ground control information is available (datum
problem),
— Not enough deformation control information is available
(degree of polynomial problem due to over-
parameterisation),
— Bad distribution of ties respective high correlation between
adjacent strips due to weak ground control (typical
polynomial oscillations).
ORIENT has built in a regularisation on the fly; i.e. when a
singularity occurs (solving the normal equation system), a
fictitious observation for the affected unknown will be
generated allowing the decomposition process to continue. This
is done automatically — the user is informed via protocol to let
him make up his opinion about the validity of the results.
We have also to take care of getting rid of wrong hypotheses
ü,,b,,C, - 0 or W,,P,,K, = 0: Gross error detection by
data snooping is used for that. Testing of significance of the
d;,D;,C;. (0,,Q,, K; , and c, ,,c,, is also a must.
4. MINIMAL DISTRIBUTION OF GROUND CONTROL
POINTS
We suppose that Lidar-strips have a similar geometric
behaviour as strips in DGPS-supported aero-triangulation. We
have to cope with deficiencies of the kinematic GPS as drift and
even jumps on turns. In the meanwhile — as long as no
exhaustive tests (simulations) are performed we suggest ground
control to overcome the phenomena. The background of the
following figure 3 is discussed in (Kraus, 19972).
€ Ur m
hr] e e 5 iP = e
| 4 | {
I d | r 7
bos | i f
[d ij
KA | I
F1 | ul
i { i i i
|| | |
A i i
£s | * * pn i
F1 | 171
i i | | i
| | M M oM
| | Pal
i | ; i i
b | à » © i & + A
ool ®t BJ 6 II UT ue TI
« Height Control Point
x for Datum
à Fuil Control Point
& e Conlrol Point for 2nd degree
Figure 3: Minimal distribution of ground control points
5. BLOCK MONTAFON
This block, covering Gargellental and Garneratal in the region
Montafon of Vorarlberg, stretches in altitude from 880m to
2875m, so spanning 2000m in height extent. So, this block had
to be flown in two missions, one of them covering the valley
regions with 24 strips the other one the superior areas with 52
strips. 4 of the 24 were cross-strips, and 3 of the 52.
559
Mainly in the crossing strips tie positions were selected
according figure 2 and then plane patches were searched for in
every overlaying strip automatically. Acceptance criteria for tie
patches were: more than 12 points with a standard deviation
less than 5cm from the adjusting plane. Since the flown data
had been clipped by the vendor at the project limits, a lot of
strips lost their crossing partner. For these strips extra tie points
had to be determined. Altogether, 1002 such plane-patches were
used; the many, 340 of them occurred in 5 strips, 6 of them
even in 15 strips, but also 244 only in 2 strips. Only 4 patches
showed up as mismatch and had to be evicted by error detection
methods. Additionally, the LVA Feldkirch hat prepared 42
ground .patches (supported by 170 points on roofs in easily
accessible areas) in a height range from 850m to 2114m. These
control patches were found in up to 14 strips.
Moreover, 18 patches on football fields were also used as height
control. The adjustment of all these mentioned observations was
done to determine GPS-shift and IMU-misalignment of each of
the two flight missions; moreover, experiments with GPS-shift
and IMU-misalignment individually for every strip were
undertaken using preventive regularisation. The analysis of the
variants is still in progress.
6. CONCLUSIONS
Geo-referencing can be greatly improved doing sophisticated
adjustment of parameters based on a manifold of hybrid
observations (GPS, IMU, laser scanner, tying planes, and
ground control). Height corrections alone do not suffice. 3D-
correction of exterior as well as interior orientation and
calibration parameters is necessary.
For high demands in accuracy — not mere precision — we need
some ground control. The ideal configuration of control points
is not yet known. With high probability the same procedure as
used for GPS-supported aerotriangulation (Kraus, 1997a,p157,
fig B5.3-5) can be recommended: i.e. control points in the
corners of a block together with cross-strips at the ends of the
block. These cross-strips may be replaced by chains of height
control points at the ends of the block.
The area of interest should be extended by about one strip-
width to grant consistency of the strip-sewing .
Quality control of a block is necessary: graphic representations
of discrepancies is a must to detect any system anomalies.
REFERENCES
Burman, H., 2000, Adjustment of laser scanner data for
correction of orientation errors, IAPRS Vol. XXXIII, Part B3.
Amsterdam 2000.
Cramer, M., 2000, Genauigkeitsuntersuchungen zur GPS/INS-
Integration in der Aerophotogrammetrie, Dissertation, Fakultät
für Bauingenieur- und Vermessungswesen, Universität Stuttgart
Heitzinger David, 1996, 3D-Oberflächenmodellierung mit
topologischen Grundelementen, Diplomarbeit IPF TU Wien
(Begutachter: Kraus / Betreuer: Kager).
Kager, H. ,1995, ORIENT, A Universal Photogrammetric
Adjustment System, Reference Manual VI.7, Institute of
Photogrammetry and Remote Sensing, TU Vienna.
Kager, H., Kraus, K. , 2001, Height Discrepancies between
Overlapping Laser Scanner Strips. In Grün/Kahmen (Eds.):
Optical 3-D Measurement Techniques V, pp. 103 -110.
