ISPRS Commission III, Vol.34, Part 3A „Photogrammetric Computer Vision“, Graz, 2002 
estimated parameters for shift, rotation and scale, the 
approximate image coordinates of the visible object model as 
depicted in Figure 2 can now be improved. By application of 
these corresponding points in object and image space, the 
original exterior orientation can be improved by spatial 
resection. Figure 4 shows the 3D building model projected to 
the image based on the refined orientation from this process. 
  
  
Figure 4: Improved mapping of the building model 
In principle, the complete process - extraction of building 
silhouette, improvement of image coordinates by GHT and 
spatial resection — has to be iteratively repeated in order to 
avoid errors resulting from the simplification of the original 3D 
to 2D matching to a 2D to 2D problem. Nevertheless, for our 
application the differences between the projected wire-frame 
and the image were mainly caused by errors within the 
available model due to measurement errors or generalisation 
effects. Thus this iteration was not applied. 
  
Figure 5: Additional example of shape matching. Silhouette 
from direct measured exterior orientation is given in black, re- 
fined localization is given in white. 
Figure 5 gives an other example of the matching process. The 
silhouette of the visible building as derived from the available 
3D model and the directly measured parameters of exterior 
orientation is depicted in black, whereas the of the image based 
refinement is outlined in white. 
The detection of the overall shape of buildings as it is 
demonstrated by these examples of course requires a sufficient 
distance of the user to the depicted building. Since we are 
aiming on applications like telepointing or navigation, this can 
be presumed for most cases. Additionally, an area covering 
provision of position information within a complex urban 
environment is only feasible based on hybrid systems. Thus in 
our opinion different sensors and techniques and sensors will 
be applied for different user scenarios. As an example, tagging 
techniques can be employed for objects of interest, which are 
already used for indoor applications. By these systems object 
identification and location is realised by a tag fixed to an object, 
which sends an unique ID i.e. via infrared signals. 
In addition to the access to object related information, precisely 
georeferenced terrestrial images in urban environments can also 
be applied for the refinement of available 3D city models. 
Whereas building geometry can usually be provided effectively 
from airborne data, the collection of facade geometry including 
doors, windows and other unmodeled structure as well as the 
extraction of facade texture currently is a current bottleneck 
during data collection. Thus, the automatic alignment of 
terrestrial images as it is feasible by our approach is an 
important prerequisite to facilitate an efficient enhancement of 
existing 3D city models. 
4. GENERALISATION OF 3D BUILDING MODELS 
During personal navigation, which is one of the main tasks 
within location based services, the visualization of the 
environment and the generation of a virtual walk through for 
planning of actual tours are features of great importance. Due to 
the small displays of a mobile device, the amount of 
information to be presented has to be reduced for this purpose. 
Hence, an automatic generalization of the 3D building models 
to be presented to the user has to be made available. 
Since a building representation by planar faces and straight 
edges is feasible for most cases, the reconstructed buildings are 
usually described by general polyhedrons. Hence, for real-time 
visualization the number of faces to be displayed for each 
building object has to be reduced considerably. In general, this 
process presumes the elimination of unnecessary details, 
whereas features, which are important for the visual impression 
have to be kept. Especially for man-made-objects like 
buildings, symmetries are of major importance. For this reason, 
during the process of generalization the preservation of regular 
structures and symmetries like parallel edges, perpendicular 
intersections or planar roof faces has to be guaranteed. 
In our approach a simplification of polyhedral building models 
is achieved by combining techniques both from cartography and 
computer graphics. In cartography a lot of effort has already 
been spent on the generalisation of 2D building structures. 
(Sester 2000) for example uses least squares adjustment for the 
generalization of building ground plans. Approaches for 3D 
object generalization have only be proposed recently (Mayer 
2000). On the other hand, surface simplification is a widely 
used technique in the field of computer graphics in order to 
speed up the visualization of highly complex 3D models 
(Heckbert & Garland 1997). Usually, surface simplification is 
applied to general objects, which are either given as polygonal 
or as triangular surface meshes. Usually the elimination of 
edges for object simplification is only controlled by geometric 
properties. Symmetry considerations, which are important for 
the visual impression of objects like buildings are not taken into 
account. These symmetries and regularities are stringently 
preserved during generalization by our approach by integration 
of a set of surface classification and simplification operations. 
The initial step of the generalisation algorithm is to build the 
so-called constrained building model, which represents the 
regularization constraints between two or more faces of the 
polyhedral building model. In the following steps the geometry 
of the constraint building model is then iteratively simplified by 
A - 141