rbul 2004 
HES 
performed 
F707. The 
Is ca 34 
h unique 
JNESCO. 
nistry of 
:d (Figure 
h a self 
ints semi- 
ig (LSM) 
ns of the 
„= 0.021 
1ce stood, 
| imported 
IT, 1999]. 
measured 
measured 
oints was 
red points 
  
   
  
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B5. Istanbul 2004 
+ on 
i i SRI ci 
po es 88 À e 6 
Figure 13: The recovered positions of the cameras and the 
measured points for the modeling of the empty niche of the 
Great Buddha. 
  
The surface generation was then performed with a 2.5 Delaunay 
method, dividing the measured point cloud into separate parts. 
A mesh was created for each single point cloud and then all the 
surfaces were merged together with Geomagic Studio 
[Raindrop]. The final model of the empty niche is displayed in 
Figure 14. 
  
  
  
  
  
Figure 14: 3D model of the empty niche of the Great Buddha, 
visualized in shaded (left) and textured mode (right). 
The recovered 3D model of the empty niche allows the 
comparison between the actual situation and the previous one, 
before the destruction (Figure 15 and 16). 
   
  
    
e TES A — B octets d -— i 
Figure 15: The niche of the Great Buddha: a comparison 
between an image of the 1970s and the actual situation. 
+ 
© 
    
   
  
B & S mde 3 zb. 
Figure 16: The reconstructed 3D models of the Great Buddha 
and the actual empty niche. An image of the full 3D model of 
the Great Buddha is shown in Figure 2. 
4.2 The empty niche of the Small Buddha 
The modeling of the empty niche of the Small Buddha was 
performed using 9 images (Figure 17) 
   
    
: TUNE aes ados 
Figure 17: The empty niche of the Small Buddha. 
The necessary tie points were measured semi-automatically with 
LSM and then imported in a self-calibrating bundle adjustment. 
The final average standard deviations of the object coordinates 
are 0, = 0.015 m, 6, = 0.019 m, 0, = 0.022 m. 
  
Figure 18: The camera poses and the manually measured points 
for the modeling of the empty niche of the Small Buddha. 
Afterwards manual measurements were performed in VirtuoZo 
on the distortion-free images and a cloud of approximately 17 
000 points was recovered (Figure 18). 
For the mesh generation, we had to split the point cloud in 
different parts, to be able to perform 2.5 Delaunay triangulation.