21 
the outlines of a 
il photos, whereas 
:. Using the width 
rue position of the 
; real height of the 
1, 2007). For many 
Remote sensing is 
are unknown. Any 
ition about the 3D 
ed as a multi-step 
application and the 
terrain or the 3D 
use the time differ- 
) shape of the area 
fusion is based on 
generated by stan- 
2trie SAR, LIDAR 
restrial data acqui- 
immetry or video- 
s possible. Further 
ration of 3D mod- 
s not yet been pre- 
ig between the im- 
ase, the 3D shape 
1 earlier. Changes 
- can be detected 
CHANGE 
or terrestrial data, 
automated gener- 
HS footprint infor- 
renner, 1999). The 
;d using this auto- 
, some models are 
ding models 
ed for change de- 
Dccur. 
irlsruhe (left) and 
ht) 
In Figure 11 the real DOSAR image and the SARViz simulation 
of the erroneously reconstructed test area are depicted. DOSAR 
is the multi-frequency polarimetric airborne SAR system of the 
EADS Domier GmbH (Hoffmann & Fischer, 2002). The off-na- 
dir angle is 70° and the 3dB-resolution is about 0.57m. For test 
ing purposes an artificially changed model, depicted in Figure 
12, has been used. The black building has been added to the 
model to test the proposed change detection algorithm. 
Figure 12. Changed 3D model 
The detected changes are visible in Figure 13. Not only the 
changes from the artificially inserted house are visible, but addi 
tionally some changes in the centre of the building block are 
visible and even larger differences are detected in the south-east 
comer of the building block. These changes are directly related 
to the incomplete and erroneous city model illustrated in Figure 
10. 
Figure 13. Detected changes of the building block depicted in 
Figure 12 overlaying a DOSAR image 
As already presented before (Balz, 2004), the false alarms are 
outweighing the detected changes. For automated change detec 
tion based on simulation assisted fusion of automatically gener 
ated 3D data, the building reconstruction process is crucial. 
More reliable building models can be reconstructed using semi 
automatic reconstruction systems. The “CyberCity Modeler” 
published by Griin and Wang (1998) is commercially available. 
Using the semi-automated reconstruction the false alarms based 
on wrongly reconstructed models will be reduced. Due to gene 
ralizations in the reconstruction process and objects which are 
not included in the reconstructed building models, e.g. trees or 
parking cars, still various false alarms will occur. 
Semi-automated change detection or alarm verification proces 
ses are another way of reducing false alarms. Assumed changes 
can be presented together with the model and the real image, 
clarifying the reason for the assumed change. Errors in the 
building model as well as possible false detections due to the 
object environment can be analyzed fast and reliable by human 
interpretation. 
Figure 14. Original DOSAR image (left) and SARViz simula- 
tion (right) of the modified 3D model from Figure 
12. 
Comparing the real image and the simulation separately, errors 
and changes are not easily recognizable. For example in Figure 
14, the erroneous building added to the model cannot be found 
easily. In Figure 15, the simulation is overlaying the real SAR 
image in real-time. The transparency is selectable and the simu 
lation parameters can be changed at any time. Using this way of 
data visualization, changes can be detected easily and reliably. 
Figure 15. SAR image overlaid by a SARViz simulation 
Changes and effects of assumed changes on the acquired SAR 
or optical image can be analyzed in real-time using visualization 
applications. False alarms due to erroneous building models can 
be detected, for example, by visually comparing aerial images 
of the area overlaid by computer visualizations of the building 
models. This way of false alarm detection and quality control is 
depicted in Figure 16. 
Figure 16. Comparison of the aerial image (left) the image over 
laid by the model (middle) and the oblique view of 
the building model (right) 
Interactive and simulation based change detection is a reliable 
way of analyzing SAR data in urban areas. Beside the detection 
of changes, the dimension of the change can even be measured 
by interactively adapting the shape of the model and visually 
comparing the simulation of the adapted model and the real 
SAR image. Thanks to the real-time simulation capability, this 
can be done fast and user-friendly.