ul 2004
eft) and
e build-
ns, or if
| Is nec-
ig block
son in a
10t nec-
In this
d to de-
ards be
the real
indicate
v-values
d image
ver and
object.
t of the
laying a
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004
In Figure 16, the automatically detected changes of the building
block can be seen. Obviously not only the artificially imple-
mented changes are visible. Furthermore, some objects inside
the building block, are not modelled correctly in the 3D-city
model. The biggest change is visible at the south-east corner of
the building block. The roof of this building is wrongly shaped
in the city model (see Figure 17).
pe:
C
Figure 17. Orthoimage (left) and erroneous shaped 3D-model
(right)
The errors and the incompleteness of the 3D-city model results
in a lot distortions in the change-detection process. The real
SAR image differs in many ways from the simulated image. For
example, objects not contained in the model, may interfere the
analysis. In Figure 13, this problem can be seen. The layover
area of the building is influenced by the backscatter from the
kerb near the building. In this case, those areas are overlapping
and this is no problem. For an even taller building, the kerb in
the middle of the street would have merged with the layover
area, indicating a much larger layover area and disturbing the
analysis.
Altogether the change detection based on SAR simulated im-
ages of 3D-models works pretty well using high-quality 3D-
models. The main problem of change detection applications
using SAR and other types of data is the geo-referencing. By
simulating the 3D-models and using the simulated image as
basis for the geo-referencing, a highly-accurate geo-referencing
is possible. The change-detection works well after the success-
ful geo-referencing, but still errors in the model itself are
remaining. For example, the detected changes in Figure 16 are
mainly based on errors in the used 3D-city model.
6. CONCLUSION
A prerequisite of every change detection operation is the geo-
referencing of the different datasets. Combining SAR data with
data acquired by different sensors, the geo-referencing of these
different data types is problematic. SAR systems are side-
looking systems with run-time geometry. Therefore the SAR
geometry of SAR images differ a lot from optical images or GIS
datasets. The different imaging properties have to be consid-
ered.
A SAR simulator can be used to transform the 3D-models into
the SAR image space. Comparing the real SAR image and the
simulated SAR image of the 3D-model, a meaningful change-
detection is possible. This comparison can be used to automati-
cally geo-reference SAR images to SAR simulated images of
3D-models. Using GDF-street data the initial geo-referencing of
the SAR image could be improved from an offset of about
150m to around 6m. Using 3D-models the offset could be re-
duced to 1.5m, although the used model was quite erroneous.
For a successful change detection using SAR images, it is most
useful to rely the detection on 3D-data instead of using 2D-data.
477
The side-looking property of a SAR system makes it most im-
portant to regard the 3D-shape of the analysed object. The
simulated image of the 3D-models, created by a SAR simulator,
can be compared to the real SAR image and as a result of this
comparison changes may be detected automatically. The final
result of the change detection depends on the quality and com-
pleteness of the 3D-model simulated for the comparison. Fur-
thermore errors in the real image and during the SAR simula-
tion can disturb the result of the change detection.
The side-looking sensor principle of SAR is unfavourable for
urban environments, compared to aerial imagery or LIDAR.
Unfortunately in those environments any detection can be pre-
vented by occlusions and disturbed by ambiguities. Under some
circumstances no change detections is possible at all, using
SAR. For time-critical applications SAR is anyhow still the best
alternative, especially during the night or under bad weather
conditions.
7. ACKNOWLEDGEMENTS
We thank the EADS Dornier GmbH for providing us with the
simulated and real SAR data and for their help in making this
work possible.
REFERENCES
Ender, J.H.G., Brenner, A.R., 2003. PAMIR - a wideband
phased array SAR/MTI system. In: /EE Proceedings - Radar,
Sonar and Navigation, 150 (3), pp. 165-172.
Haala, N., Brenner, C., 1999. Extraction of buildings and trees
in urban environments. In: ISPRS Journal of Photogrammetry
and Remote Sensing, 54 (2-3), pp. 130-137.
Hoffmann, K., Fischer, P., 2002. DOSAR: A Multifrquency
Polarimetric and Interferometric Airborne SAR-System. In:
2002 International Geoscience and Remote Sensing Symposium
and the 24th Canadian Symposium on Remote Sensing,
Toronto, Canada.
Leberl, F.W., Bolter, R., 2001. Building reconstruction from
Synthetic Aperture Radar images and interferometry. In:
Baltsavias, E.P., Grün, A., Gool, L.v.: Automatic Extraction of
Man-Made Objects From Aerial and Space Images (III). Lisse,
pp. 281-290.
Levine, M.D., Shaheen, S.I., 1981. A modular computer vision
system for picture segmentation and interpretation. In: /EEE
Transactions on Pattern Analysis and Machine Intelligence, 3,
pp. 540-556.
Sórgel, Uwe, 2003. /terative Verfahren zur Detektion und
Rekonstruktion von Gebdáuden in SAR- und InSAR-Daten. PhD-
Thesis, University of Hannover.
Sôrgel, U., Schulz, K., Thoennessen, U., Stilla, U., 2003.
Event-driven SAR Data Acquisition in Urban Areas Using GIS.
In: GIS, 16(12), pp. 32-37.
UNCHS, 2001. The State Of The World's Cities Report 2001.
United Nations Center for Human Settlement, Nairobi, Kenya.
Walter, V., 1997. Zuordnung von raumbezogenen Daten - am
Beispiel ATKIS und GDF. Deutsche Geodátische Kommission
(DGK), Munich.
