n the 
najor 
s are 
xture 
Iding 
tion, 
1 for 
ented 
rious 
ction 
etical 
les, a 
cking 
ction 
own, 
5). In 
cause 
asses 
luded 
other 
jects 
s can 
netric 
Itural 
imple 
> also 
even 
o the 
other 
ts. In 
ry to 
other 
n the 
vhich 
ess is 
  
Bofang Zhao 
described in Section 3, followed by building extraction in Section 4. Results of tests are given in Section 5 and a 
conclusion is drawn in Section 6. 
2. OVERVIEW 
In this research, a building is modelled as a polyhedron, comprising planes that are connected to form a solid volume. 
The intersections of adjacent planes are straight lines. Furthermore, the polyhedron has a set of attributes and 
knowledge describing its geometry, radiometry, texture, topology, and context. From this model and for attacking the 
complexity of the problem, the designed system consists of three parts, namely building detection, building segment 
extraction and 3D segment matching and building modelling. Within this hierarchical framework, the processes of 
building detection and reconstruction are purposely separated to achieve greater reliability and efficiency. For the same 
reason 2D and 3D procedures are applied mutually cooperatively. Hence, buildings are hierarchically reconstructed 
from global detection to local modelling, that is, from coarse to fine details. In addition, all components of the system 
are modular, thus allowing flexibility and extendability of the system and components. Figure 1 shows a block diagram 
of the system. 
3D Buildings 
  
Building Modelling agg — — — —, 
^ 
— Feature Matching 
1 
Building Detection 
A 
  
  
Building Segment 
Extraction 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
DSM Segmentation 
  
  
  
Shadow Extraction A Texture Analysis 
  
  
  
  
  
  
  
DSM Acquisition 
A 
  
  
  
      
Domain, Context 
Knowledge 
Figure 1. Block Diagram of the System 
The detection process starts with segmentation of DSM (digital surface model) to derive regions of interest (ROIs) 
which have high expectancy of representing individual buildings. Texture and shadow information are extracted and 
used to refine and verify ROIs. All subsequent reconstruction procedures are focused locally on those detected building 
regions, thus greatly reducing the complexity of reconstruction process. Buildings are constructed in a bottom-up 
approach. Primitive linear features are first derived, and relevant building segments are extracted by grouping and 
filtering these primitive features within individual building regions. 3D lines are then generated by feature matching of 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 1027