Extracted weak 
edge. 
  
Extracted edge 
partly occluded 
by trees. 
  
lding boundary. 
of multiple roof 
of surface, or its 
igure 6 through 8 
ire 8(b) shows an 
of this algorithm. 
  
ISTINGI 
USE ' 
“built building 
=non-candidates) 
1 
> filter 
ge 
esponding to the 
tree, road, etc.) 
ons by region 
ion combinations 
| are the most 
lidates 
ns list for each 
ype in Figure 5) 
building model 
e edge and the 
lfillment rate by 
el instances in 
each instance; 
region merging 
istances, Table 1) 
ctangular fitting, 
e (« 0.4) 
» information 
n algorithm 
  
  
        
Edge selection 
in the orthogonal Os 
direction to S 
the sun vector 
    
Candidate 
Candidate edge C 
region = 
Initial 
: rectangle 
Candidate VAN d 
edge A^ / Initialize instances 
for each candidate 
edge 
  
  
  
Candidate region 
Step 2.1: Building candidate region generation 
Step 2.2: Building hypotheses generation 
Edge at 90 dgrees Rectangular Rectangular 
Edge at 135 building model building model 
degree 
INSTANCE A 
   
   
   
  
    
    
    
   
    
   
INSTANCEC # 
   
Rectangular 
building model 
  
  
  
  
Step 2.3: Optimal 
building hypothesis search 
Figure 7. Newly-built building extraction 
/ 
I 
\ 
“._INSTANCEA _~ 
z 
^ Rectangular ^* 
// Rectangular ^* 
‘ 
buildingmodel } |. buildingmodel } 
Connection 
Repeat until 
convergence 
Evaluate the result 
M ui and finalize process 
“.._INSTANCEC _~ 
      
Unification 
Rectangular 
building model 
pert Extracted 
/ building 
  
  
  
  
  
Step 2.4: Post-processing 
  
  
Rule type Inclusion Connection Unification 
Geometry contain / inside / covers / equal touch / adjacent overlap / intersect 
relationship 
  
* Merged to the one which |» 
includes (the other) 
Merged to the any when the 
range is equal 
Merge rules 
Merged to the BRIGHT dominant one 
only when the rectangular fulfillment 
rate improves 
Merged to the one with a higher 
rectangular fulfillment rate 
Merged to the one with a longer 
edge when rectangular fulfillment 
rate is the same 
  
  
Examples 
  
  
  
BRIGHT 
dominant 
combination: 
1 
  
  
   
  
MIDDLE 
dominant 
combination 
  
TU 
   
  
  
  
  
Generation of Candidate Candidate selection 
convex full rectangular for the highest 
and candidate rectangular 
rectangular fullfillment rate 
Convex hull 
of candidate 
region Extracted 
  
Building Candidate region 
edge adjacent tothe edge 
   
   
   
Foreach . 
. 
combinations ? 
: 
newly-built 
Candidate region building 
excluded from 
the combination 
(Best fitted 
rectangular) 
Figure 8. Rectangular fitting. (a)Algorithm. 
(b)Example of rectangular fitting result (5 or 41 regions selected, rectangular fulfillment rate 0.45). 
73 
  
(b)