'. Istanbul 2004
est area).
ine areas.
| »
.0 m m
"st area).
pine areas.
ts. We compute
from our DSM
accuracy of the
depends on the
! open areas. In
se. The analysis
s are almost all
reas (with some
an 70 percent of
he results show
nt definitions in
so, the different
laserscans may
1., 2004).
n meter
DEM Size Height
Accuracy
5km x 5km 05
5km x 5km 05
5km x 5km 05
5km x 5km 05
Okm x 1.3km 05
Okm x 7.7km 5.0
50km x 30km 20
aset *DLR-DEM-
ter grid)
many 12
1e ISPRS-CNES
S stereo images.
aria and a part of
Table 4 gives
air from SPOTS-
je morning. The
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
images have the resolution of 5 m in along-track and 10 m in
across-track directions.
The test area includes a mountainous area (rolling and strongly
inclined alpine area) and hilly areas (rough/smooth and weakly
inclined areas). Our image matching software not only generates
a large number of mass points, but also produces line features.
The TIN based DSM was generated from the matched mass
points and the edges (as break-lines).
Figure 8 shows the 3D visualization of the generated DSM. The
results show that the shapes of our generated DSMs are similar to
the references, but slightly smoothed. This can be expected
because of the 5m resolution of the satellite images.
Tables 5 and 6 show the DSM accuracy test results. The
orientation accuracy is about 6.3 m in planimetry and 2.6 m in
height. We compute the differences between the heights of the
reference DEM and the interpolated heights from our DSM.
Table 6 shows the DSM accuracy test result by masking out the
tree areas manually.
From Tables 5 and 6 it can be seen that:
o The accuracy of the generated DSM is more or less at the
Ipixel level or even better. Only the datasets 5 give values at
about 2 pixels, but these higher values are mainly caused by
frees.
o All datasets still contain some blunders, which our procedures
failed to detect.
e The results show systematic errors. In datasets 5-1 and 5-2 the
biases are about 1 pixel. Except in case of dataset 6 all biases are
significantly negative. This indicates that our generated DSMs
‘are higher than the reference DEMs, an effect which could be
expected..
Table 5: DSM accuracy, units are meter
verage
Difference | Difference
Table 6. DSM accuracy, units are meter
excluding the tree covered areas
verage
Difference | Difference
5. CONCLUSIONS
In this paper we have reported about our current matching
approaches for fully automated DSM generation from linear
array images with different resolutions. We have developed a
matching strategy combining feature point matching, grid point
matching with neighborhood smoothness constraints, and robust
edge matching. The strategy allows us to bridge over areas with
little or no texture and at the same time maintain the important
contribution of object/image edges. The modified MPGC is used
to refine the matching results in order to achieve sub-pixel
accuracy. The geometrical constraints are derived from the
specific sensor models for the linear array imagery, which can be
the rigorous sensor model for aerial and satellite images or the
RF (Rational Function) model for satellite images.
As evidenced by a visual inspection of the results we can
reproduce even small geomorphological features. The results
from the quantitative accuracy test indicate that the presented
concept leads to good results. If the bias introduced by trees and
buildings is taken out, we can expect a height accuracy of one
pixel or even better from satellite imagery (e.g. IKONOS and
SPOT) as “best case” scenario. In case of very high resolution
aerial images (footprint 8 cm and better) it is obvious that the
“one pixel rule” cannot be maintained any more. Alone surface
roughness and modeling errors will lead to large deviations, such
that an accuracy of three to five pixels should be considered a
good result. This is at the same level as laser scanning results. Of
course, the photogrammetric data can also be produced with the
same or even better point density. On the other hand, with these
accuracies we are still operating at a coarser level than with
manual measurements from analogue aerial images, but we do
that with the advantage of great gain in processing speed.
A major problem left is the control and automated detection of
small blunders, which still infest the results, despite the
simultaneous matching of more than two images. This
constitutes a relevant topic for further research.
ACKNOWLEDGEMENTS. We appreciate the support of
the Swiss Federal Office of Topography, Bern, which provided
the laserscan data. We also thank Henri Eisenbeiss, who helped
in setting up the Thun area as a testfield for highresolution
satellite image processing.
REFERENCES
Baltsavias, E. P., 1991, Multiphoto Geometrically Constrained
Matching. Dissertation, IGP, ETH Ziirich, Mitteilungen No.
49, 221 pages.
Eisenbeiss, H., Baltsavias, E., Pateraki, M., Zhang, L., 2004.
Potential of IKONOS and QuickBird Imagery for Point
Positioning, Orthoimage and DSM Generation. /APRS, Vol.
35, Istanbul, Turkey (to be published)
Gruen, A. 1985, Adaptive Least Squares Correlation: A
powerful Image Matching Technique. South Africa Journal of
Photogrammetry, Remote Sensing and Cartography, 14 (3),
pp. 175-187.
Gruen, A, Bàr, S, Bührer, Th, 2000: DTMs Derived
Automatically From DIPS - Where Do We Stand?
Geoinformatics, Vol.3, No.5, July/August, pp. 36-39.
Gruen, A., Zhang, L., 2002. Sensor Modelling for Aerial Mobile
Mapping with Three-Line-Scanner (TLS) Imagery. /SPRS
Commission Il Symposium on Integrated System for Spatial
Data Production, Xi'an, P. R. China, August 20 — 23.
Gruen, A., Zhang L., 2003. Automatic DTM Generation from
TLS data. Optical 3-D Measurement Techniques VI, Vol. 1,
Zurich, pp. 93-105.
Kanade, T., Okutomi, M., 1994. A Stereo Matching Algorithm
with an Adaptive Window: Theory and Experiment. /EEE
Transactions on PAMI, Vol. 16, No. 9, pp. 920-932.
Murai, S., Matsumoto, Y., 2000. The Development of Airborne
Three Line Scanner with High Accuracy INS and GPS for
Analysing Car Velocity Distribution. ZAPRS, Vol. 33, Part
B2, Amsterdam, pp. 416-421
Otto, G. P., Chau, T. K. W., 1988. A "Region-Growing"
Algorithm for Matching of Terrain Images. Proc. 4^ Alvey
Vision Club, University of Manchester, UK, 31 Aug. — 2
Sept.
Poli, b. Zhang, L., Gruen, A., 2004. SPOT-5/HRS Stereo Image
Orientation and Automatic DSM Generation. ZAPRS, Vol 35,
B 1, Istanbul, Turkey (to be published)
Saint-Marc, P., Chen, J-S., Madioni, G., 1991. Adaptive
Smoothing: A General Tool for Early Vision. [EEE
Transactions on PAMI, Vol. 13, No. 6
Prazdny, K., 1985. Detection of binocular disparities. Biological
Cybernetics, Vol. 52, pp. 93-99.