Although the last ten years have seen the 
development of applications of 3D urban data 
bases (essentialy for urban development or 
mobile radio networks planning), the growth of the 
market has been slow. Two bottlenecks have been 
identified by today’s end-users: 
e quality-cost ratio is too low, especially 
because of the labour-intensive creation of the 
3D data base (including the creation and 
processing of photorealistic texture); 
«limited accessibility of the data base by end- 
users, essentialy due to the lack of a structure 
capable of organizing the data and capable of 
providing network access to remote users. 
Both bottlenecks need to be overcome, if the 
broad acceptance of fully threedimensional data 
bases of urban areas shall arise. The need of 
advanced techniques to acquire source data, 
create photorealistic textured models of a citiy's 
buildings and objects and to maintain and 
distribute these data is evident. 
Geometry vs. Texture 
The photo-realistic rendering of CAD models from 
real-world objects, e.g. the buildings and 
structures of a city, is a very current topic since a 
high degree of naturalism of a computer model is 
highly desireable. Such naturalism is needed to 
create broad appeal for digital 3-D graphics 
solutions. In the case of urban environments the 
need for so-called photo-realistic city models is 
evident from numerous applications such as 
urban planning, architecture, entertainment, 
disaster preparedness etc. 
We have now investigated the benefit of texture 
and the relations between texture and geometry. 
We know that the two sets of data need to 
correspond even in case of a multi-level-of-detail 
presentation of the computer model on each 
different level of the visualization. In [Gruber et 
al, 1995b] we have presented ideas and a flow 
diagram of a building box and roof modeling 
procedure, which points out the importance of 
correspondence between geometry and 
phototexture. 
The need of photo-texture in the threedimensional 
digital model of a citiys environment shall be 
documented within the following set of figures. A 
simple building from the Vienna City Block (see 
Fig 1) is presented as wireframe model (Fig.2 a), 
surface model using different colors for roof and 
facade (Fig. 2b) and in a more photorealistic 
manner, exploiting the texture of photographs (Fig. 
2c) The quality of the different graphics 
representations is clear, the range of usabilitiy 
may also be easily understood: 
262 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996 
  
Fig. 2: Different visualization types of the city block in Vienna; 
left) wireframe representation, middle) surface model and 
right) photorealistically textured model 
e the wireframe model seems to be 
unacceptable (even the conventional 2D CAD 
model may be easier to read and manipulate); 
e the surface model gives a coarse idea of the 
buildings’ form and may help in some special 
applications (large area city planning etc.); 
e the phototextured model allows an immediate 
identification of the unique building; the data 
set is the basis of digital visual information, 
which is easy to understand and supports the 
human operators visual sensitivity; 
e the fusion of manipulated texture and the 
detailed geometric model based on primitives 
also leads to an inadequate, unrealistic 
impression considering sharp and unnatural 
edges; so the extension to a combined 
manipulation of geometry and texture at the 
intersection of primitives has to be considered 
to improve the visual impression. 
We now argue that photorealistic texture is an 
essential part of city models. Only this high 
degree of realism will also open the digital model 
to a broad useabilty in the growing markets of 
multi-media and tele-services. 
Towards Automation 
Creating 3-D models of cities needs a large 
amount of manual processing. The transition of 
existing data like GIS or DEM, the modelling of 
buildings and other objects which are not yet 
available in three dimensions and the acquisition 
of missing data like texture from facades of 
buildings are performed under time and manpower 
consuming circumstances. Therefore we need to 
automate these processes on a dramatical scale 
[Leberl et al. 1994], [Gülch 1992], [Lang et al. 
1993]. This automation shall take place on three 
different levels of the modelling procedure: 
  
     
   
   
  
  
   
   
  
    
  
   
   
    
   
   
    
    
    
  
    
   
   
    
  
    
    
  
  
  
   
   
   
  
   
    
    
      
  
   
   
   
    
    
e 
«O0 DD =" = 
— 
The 
and 
qual 
acce 
rese 
Con 
In lir 
abo 
work 
func 
mod 
conc 
the 
Esp! 
distr 
pow 
pres 
The 
ge— — Sm P 0 p 
digi! 
pro 
Cy 
con