International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004 
  
  
  
Fig. 6: a, b) slant range magnitude (a) and phase (b) image 
overlaid with footprints transformed using mean building 
height; b) simulation result based on building footprints and 
mean height; c) simulation result based on LIDAR DEM; d) 
image a) together with detected line scatterers (yellow); e) 
LIDAR DEM, map and line scatterers (yellow). 
504 
5.4 3D city model 
In order to address salient line scatterers a vector representation 
of the object planes is required. Three dimensional city models 
provide such information. Here, the buildings were 
reconstructed using the LIDAR DEM and the building 
footprints [Stilla and Jurkiewicz, 1999]. 
From this vector data, possible locations of double-bounce 
effects and specular reflection are identified. In the first case, 
vertical building planes oriented towards the sensor are 
segmented. Building planes with a normal pointing to the 
sensor cause specular reflection. In both cases, occlusion from 
other objects in front is considered. The detected structures at 
building walls and roofs, e.g. the salient superstructures on 
buildings B and J, are shown in Fig. 6f. They match well with 
bright lines in the acquired data (Fig. 6e). The image interpreter 
may benefit in many ways from this kind of information. For 
example, the bright line below building E is actually caused by 
a double-bounce event and therefore located at the true position 
of the building footprint. The absence of predicted line 
scatterers or the appearances of additional ones are hints to 
changes in the scene. Furthermore, the polarimetric behaviour 
of the mapped objects can be predicted from the plane 
orientations e.g. of the building structures. 
5.5 Fusion of multi-aspect SAR data 
By a fusion of multi-aspect SAR data e.g. occlusion (shadow) 
areas can be filled and layover effects can be compensated. In 
general, the data fusion might be carried out at the iconic or the 
symbolic level. In the iconic case, often the orthorectified SAR 
imagery are fused, e.g. by choosing the brightest amplitude 
pixel or the DEM value with best coherence. 
This method has some drawbacks. Firstly, if an InSAR DEM is 
used for orthorectification, straight object contours might not be 
mapped to straight lines on the ground, because of noise present 
in the DEM. Additionally, very high georeferencing accuracy is 
a prerequisite for such pixel-based fusion. Therefore, we 
recommend to carry out first the object segmentation in the 
slant range data, use this result for smoothing the InSAR DEM 
and to transform the symbolic description together with the 
related iconic texture' [Soergel et al., 2003c]. In such a way the 
mapping accuracy can be increased, by smoothing the InSAR 
DEM. For example, the maximum likelihood height estimate of 
a flat roofed building is the average of the corresponding DEM 
values. Secondly, an iconic fusion alone can hamper image 
interpretation, because object features like cast shadow areas 
might disappear. Hence, an additional fusion at the symbolic 
level seems to be advantageous. 
With respect to the fusion of context data like maps and aerial 
images with SAR imagery we prefer the transformation of the 
reference data into the different slant range geometries and to 
superimpose it on the SAR data as shown above. In such a way 
linear structures can be compared with their slant range 
counterparts independently in each image. 
6. CONCLUSION 
In case of bad weather conditions or smoke, which do not allow 
taking useful data by aerial images or LIDAR a mapping using 
SAR is still possible. The side-looking illumination by SAR 
causes inherent artefacts particularly in dense urban areas. 
Usually some parts of the urban scene remain invisible using a 
single SAR data set. An analysis of multi-aspect SAR data 
offers an improvement of the results. The SAR acquisition 
Inte 
dire 
froi 
Esp 
by 
Par 
the 
the 
env 
obj 
Det 
is t 
of 
veg 
imz 
Bar 
Ima 
and 
Dot 
An: 
Sen 
Enc 
airb 
pp. 
Frai 
199 
Sim 
110 
Gar 
exti 
Tra 
|; P 
Gui 
Urb 
Mei 
geo 
and 
Ros 
N., 
Rad 
382