sn cs 
  
  
  
  
  
shows the result of split node border. Figure 5.b, c shows the 
best-fit plane of each split sub points set by its border and TIN 
respectively. 
  
  
  
wd. ard E 
jg aes uA us M 
e e E Rar 
ps pO ST à 
E apr: ud * 
MATES ig] Tyke Ara 
AMI RES 
vele BES a 
cM 
295 Y, :l 
   
   
    
  
SE 
T P. zs 
rid od 
SE 
2 
     
    
  
  
  
  
£4 
"e, 
anre Pay 
t oe 
RH 
«SU £ A. 
eva Vr TE Terre ird 
ee 
  
  
(a) 
  
   
  
  
  
  
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004 
2.4 Merge process 
A merge process is than performed after splitting to reunite 
similar neighbouring planes. Each node in octree structure has 
26 equal or larger neighbouring nodes at most. The octree 
structure allows us to find neighboring planes easily (Samet, H., 
1990). When a neighboring plane is found, the merge process 
will first check whether their normal vectors are similar. Figure 
6.a shows the idea of checking normal vectors. However, one 
may find planes in different layers having similar normal 
vectors (Figure 6.b). To solve this situation, additional check of 
the node relation is needed. After two neighboring planes are 
found similar, the rigorous calculation will be triggered again to 
recalculate the parameters of the merged plane and to ensure the 
merge availability. After the merge process, the original lidar 
points cloud was segmented into clusters of 3D planes. Each 
merged 3D plane can be recorded as 4 plane parameters (A, B, 
C, D) and its border points. A TIN (Triangular Irregular 
Network) of each plane is also generated for visualization use. 
Figure 7 a, b, shows the result of merged plane in the form of 
border and TIN. It is obviously that the 4 separated ground 
planes in Figure 5.b was merged into a larger plane successfully 
in Figure 7.b. after the merge process. 
  
  
(b) 
Figure 6. Similarity of two neighboring nodes A and B (a) 
available for merge (b) not available for merge 
  
  
  
  
  
  
  
Table 5. Result of splitting (a) node border (b) best-fit plane 
border (c) best-fit plane TIN ; 
   
  
Internai 
  
  
A class 
extracte 
have nc 
simple 
arça, : 
orientat 
based « 
classifi 
  
  
Figu 
Some c 
based | 
The pr 
ground 
should 
differei 
lidar d. 
and a 
Hsinch 
the bas