1092 
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Voi. XXXVII. Part Bl. Beijing 2008 
The afterwards described processing chain is based on a VHR 
stereo scene. In the first step a high resolution digital surface 
model (DSM) has to be extracted. This step is very crucial for 
the quality of the result if no additional data like building 
footprints are available. 
Most approaches for an automatic generation of 3D city 
models in literature are based on a high quality DSM. For 
example (Brenner, 2000/2003) describes fully automatic 
reconstruction systems based on a high quality laser DSM. 
Also (Gamba 2005) and (Rottensteiner, 2002) use high quality 
laser DSMs. 
In contrast the DSMs which can be derived from optical VHR 
satellite imagery suffer from much errors and outliers as 
shown in Figure 1. 
Figure 1. Left: laser DSM, right: calculated DSM from Ikonos 
stereo image pair (Section 600 m x 400 m from the Munich 
scene, area of Technical University) 
Especially the estimation and extraction of building outlines 
will become tedious due to streaking effects, outliers, 
occlusions, and other artefacts in the calculated DSM. A main 
problem originates in the demand of a DSM with a ground 
resolution in the same order of magnitude as the GSD of the 
imagery. This requirement arose from the necessity to 
distinguish and characterize urban objects which requires 
DSMs with resolutions of about 1 m or below. 
Surface models generated from aerial photographs use in 
contrast about 25 image pixels for one DSM pixel. Hence such 
DSMs are much better and can already be used for the widely 
proposed building extraction approaches in the literature. 
2. DATA 
The proposed automatic processing chain uses VHR stereo 
image pairs which are best acquired in the same orbit with the 
same illumination conditions. For the presented work two 
Ikonos stereo image pairs of the cities of Athens and Munich 
were used. 
The Athens scene was acquired 2004-07-24, 9:24 GMT, with a 
ground resolution of 88 cm and viewing angles of-19.99° and 
+13.17° respectively. The scene was delivered as a level IB 
image, full sensor corrected standard stereo product in epipolar 
geometry (Figure 2) 
Figure 2. Section 600 m x 400 m from the Athens scene, left 
and right stereo image 
Figure 3. Section 600 m x 400 m from the Munich scene (area 
of Technical University), left and right stereo image 
The Munich scene was acquired on 2005-07-15 at 10:28 GMT 
with a ground resolution of 83 cm. The viewing angles of the 
forward and backward image were +9.25° and -4.45°. The 
images were available only as level 1A product, which are 
corrected for sensor orientation and radiometry (Figure 3) but 
contain no further geometric changes. 
3. PROCESSING CHAIN 
The proposed automatic processing chain consists of the 
following steps which are explained in brief in the next sections: 
1 Preprocessing of the raw imagery 
2 Creating the digital surface model (DSM) 
3 Extracting the digital terrain model (DTM)