Task Labour % Automation Workstation 
Cameras, MIS, 2D GIS 
film scanner, PAS, 2D GIS 
DPWS 
3 DPWS + 3D CAD 
4 3D CAD 
GWS 
Source data acquisition 9 3 
Source data preparation 5 5 
Photogrammetric set-u 10 3-5 
Reconstruction of objects 20 3 
Texture processing 23 
Texture mapping e 
Tuning and visualization 10 
  
MIS, = Mobile pt S 
- Pidure Archiving tem 
CPs 26 Enn rame W EN ind. DIM 
DG - 2D Geographic Information System (in 
3D CAD z C AD System s 
TPT = Texture Processing Tool 
GWS = Graphics Workstation 
  
  
Table 1: Key processes of a modelling procedure for 
Cybercities 
object reconstruction, namely the reconstruction of roofs of 
cities’ buildings. The main source data is a set of digital 
aerial images, providing multiple overlap and the possibility 
of stereo restitution. We also involve the existing 2D GIS, 
i.e. the digital map. We take in account, that existing 
2D GIS data, which are currently maintained by cities' 
administrations have to be involved because of their techni- 
cal existence and their legal charakter [Wilmersdorf 1994]. 
The fusion process we are investigating, is based on an 
affine matching procedure, which allows to relate line 
features of the map, i.e. the footprints of buildings and 
the line features derived from digital images. From this 
first solution - the overlay of the detected roof outline on 
the digital image - we proceed to detect and refine the 
shape of the roof and details like chimneys and attic windows. 
3 FUSION OF MAP AND IMAGE DATA 
The first step of our fusion process of image-based and map- 
based data consists of the classification of the resulting ob- 
jects, i.e. buildings with respect to the type of source data 
(cf. Figure 1 and points below) and to the degree of quality 
of the resulting geometrical model of building (cf. Figure 2 
and points below). 
  
  
  
ABa 
ABal 
nani 
* 
1 L 
1 i 
1 | 
1 3 I 
I z 1 
I Ba i ABal.1 1 AR 
1 : i 
I i 
i I 
i 1 
j 1 
EE Rd unn n m un LP 
  
  
  
Figure 1: Relations between buildings in GIS (group 
A) and on aerial image (group B). 
e À - all GIS buildings, 
e Aa - GIS buildings, which don't exist in aerial images, 
e B - all buildings in aerial images, 
e Ba - buildings, which don’t exist in the digital map data 
set, but can be reconstructed using parametric 3D roof 
skeleton database or using other methods, 
e  ABa - GIS buildings with reconstructed roof outline. 
eo ABal - GIS buildings with reconstructed 3D roof skeleton, 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996 
e ABal.l - GIS buildings with reconstructed 3D roof detail. 
92.2 
Figure 2: Types of reconstructed buildings. 
  
a) 
e a) - 3D building box, roof skeleton and roof detail 
[ABa1.1], 
e b) - 3D building box and roof skeleton [ABal], 
e ¢) - 3D building box [ABa], 
e d) - 2.5D building's footprint (GIS + DTM) [A-ABa]. 
Other steps are documented and enumerated within the fol- 
lowing paragraphs and depicted by figures: 
3.1 Preparation of the image source data. 
The source image data may be analog and need to be digitized 
via a film scanning system. In this case we have to chose 
the appropriate pixel size and the radiometric resolution of 
the digital data. Initial set-up data and photogrammetric 
parameters are necessary to relate the image data to the world 
coordinate system. The digital image data is now filtered with 
low-pass filter, processed by edge detection algorithm and 
converted into so-called tokensets of lines [Burns et al. 1986] 
(cf. Figure 3). 
  
VA "© IX 
Figure 3: The line-segment tokenset as input-data for 
determination of roof-skeleton. 
3.2 Preparation of the digital map. 
The existing data set of the 2D geographic information sys- 
tem of the city shall be used to extract the footprint of each 
building. We may use in addition " Number of floors of build- 
ing" as an attribute for coarse determination of roof outlines' 
elevation above the terrain surface. 
3.3 Fusion of image data and digital map (Monocular 
approach). 
The tokenset of the digital image and the building's footprint 
are now related one to another by a piecemeal affine matching 
procedure (cf. Figure 4). The idea is based on affine match- 
ing proposed by [Pinz et al. 1995] and is applicated succes- 
sive for " pieces" of coarse roof outline - one line segment and 
        
     
     
     
      
    
   
    
    
    
   
   
   
    
  
   
   
    
  
     
  
    
     
     
   
    
   
    
   
   
    
   
   
   
   
    
   
    
two neig 
to affine 
are extr: 
and the 
at least 
proxima 
the mat 
lines are 
Fi 
th 
  
3.4 T 
The res 
by the 
stereosc 
the ext 
dimensi 
This d: 
leds to 
differen 
The me 
of a su 
and sul 
proach. 
on is de 
From t 
image « 
ploiting 
lines ar 
step by 
The fir 
interac 
(7c).