International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004 
with intersections and road polygons, which are 
represented by triangles. 
  
Figure 5. Road Networks 
2.6 To Generate Building Polygons 
To generalize building polygons, there are three steps: 
clustering building polygons, aggregating building clusters, and 
simplification, in the proposed method. 
Clustering building polygons is implemented to decide which 
buildings are near enough to be aggregated. Figure 6 depicts the 
clustering procedure. Here, to avoid erroneous clustering, road 
networks should be used as constraints for separating buildings 
compulsorily. The details are as follows. 
1. Defining links between two different buildings. À 
link's length is equal to the minimum distance of two 
linked buildings (Figure 6). Here, each link begins at 
a building and ends at another one. The numbers near 
each link is the minimum distance between linked two 
buildings. 
2. Sorting all links by their lengths in ascending order. A 
sorted result is also shown on the left-down corner of 
Figure 6. 
3. Grouping links from the first (shortest) link. If one of 
two linked buildings is included in the current group, 
another building will be appended to the current 
group. Otherwise, a new group is created with the 
current link's two buildings. A new group is also 
created if a link's length is larger than the parameter 
for clustering. 
4. Repeating the grouping until all buildings are grouped. 
  
Sorted links: 
* Cluster 1 
(5: CD) 
(0: AC) 
(6: B.D) 
12- AJ). 
C EO Cluster 2 
EN 
(9: DE) 
(10: CL) 
a 
does | lnored 
(13: BF) 
(15: AE). 
  
  
  
"Threshold 
  
  
  
  
Figure 6. Link's network for clustering buildings 
With the above procedures, we can obtain the clustered results 
of Figure 6 with different thresholds for clustering buildings, 
shown in Table 1. The left-down corner of Figure 6 is same 
with the greyed areas in Table 1. 
      
   
       
  
Buildin 
B.C.D 
Cell size Cluster ID 
S 
   
  
    
   
    
   
   
D 
      
C 
EG 
A,B,C, DERG 
— NO fm Mo = | DWN [= jn | BLD | — 
Results of clustering buildings with different 
thresholds for Figure 6 
Table |. 
Aggregating a building cluster is carried out by using TIN and 
creates aggregated buildings from building clusters. Some 
aggregated buildings are shown in Figure 7. Here, the buildings, 
which are too near to road edges, were also displaced from the 
road edges before clustering building polygons. 
(a) Original building polygons 
(b) Displaced and aggregated buildings 
Figure 7. Displacing and aggregating buildings 
252 
Internatic 
LME NE 
Simplific 
polygons 
vertexes - 
buildings 
We impl 
the effec 
the origit 
The gen 
modeled 
road net 
polygons 
single o1 
building 
fi 
Ad f^ 7 
d 
JW 
E t 
4 yP" rg 
Sf s CUN Pa OF