The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B5. Beijing 2008 
982 
solved. In our work frame, firstly, a real 3D reconstruction will 
be done including the inside and outside for the whole 492 
caves and 220 km 2 places around them, to construct a general 
digital record and virtual demonstration system for researches 
of resume and rebuild, as well as for the tours instead of the 
locale. Secondly, research on the chromatic recovering and 
rebuilding for the fresco will be implemented. Through 
studying the history of the evolvement and development of 
fresco colours, the methodologies of digital image processing 
are used to recover the fresco’s real colours. Thirdly, in the 
virtual environment, the 3D Holographic Laser Projection helps 
us to realize the 3D interpretation of the histories in Dunhuang 
for the culture inherence and communication. According to the 
workflow above, this cultural relic will be conserved precisely, 
scientifically and forever. More importantly, it is possible that 
the illustrious history for more than one thousand years of the 
Mogao Caves will recur in a virtual world. 
2. DATA ACQUIRING 
2.1 Control Surveying 
The control survey is introduced, in one hand, all the absolute 
coordinates with high precision is acquired, and it can also 
build up a conversion relationship from the relative coordinates 
acquired in ways below to their geodesic ones. It covers two 
parts, one is traverse survey, and the other is photo control point 
survey on different sides of nine-stored Mogao cave. During 
traverse surveying, three traverses namely the traverse A, B and 
C are laid down and they are annexed traverse totally. Traverse 
A is composed of 4 traverse points: GPS23, Al, A2, and 
GPS15; traverse B is made up of 5 traverse points: Al, Bl, B2, 
B3, and A2; and traverse C consists of 8 traverse points: 
GPS23, Cl, C2, C3, C4, C5, C6, and GPS 15. 
Figure 1. Traverse sketch 
Artificial mark points are distributed in the different sides of the 
nine-stored Mogao Cave (Cave No.96). Some of the 
outstanding points in the building are chosen as photo control 
points to be surveyed. After computation, measurement 
accuracy is better than ± 15mm required by the Digital 
Dunhuang Project. 
2.2 Multi-viewpoint Stereo Photogrammetry 
Stereo photogrammetry is an effective method of acquiring 
scene information and has the significant advantages in 
concentrating the objects’ 3D shape and texture. Therefore, for 
some special applications, such as those requiring high 
precision in dimensions (both in shape and position 
reconstruction dimensions, and in texture mapping dimensions) 
and high quality in vision experience of objects’ appearance, 
e.g. digital archives and 3D reconstruction of ancient buildings, 
grottoes or mural, this method has many prevailing advantages. 
The principle of fixed-baseline stereo photogrammetry is based 
on the theory of aero-photography measurement and computer 
vision (Hartley et al., 2000; Li et al., 1992) . Firstly, spatial 
resection method is used to calculate the intrinsic and extrinsic 
parameters of each camera (Gao et al., 2007). Secondly, spatial 
forward intersection can be used to calculate the 3D coordinates 
of corresponding points between the stereo images. Lastly, 3D 
model can be archived by automatic triangulation 
reconstruction or by human-computer interaction from the point 
clouds, after rectifying the image and its rectified dense 
matching. Of course, wire frame model can be directly obtained 
by human-computer interaction, i.e., measuring and modelling, 
and owing to the vision based on their process, the dimension 
and position of the texture combined with its mapping are in 
high precision. 
(a) (b) 
Figure 2. (a) The image from the left camera; (b) the image 
from the right camera 
2.3 Laser Scanning 
Recently, laser scanning technology has been widely used in 
the model reconstruction for ancient buildings (Bonora et al., 
2005) . It can acquire large amount of points with high density 
from the surface of an object. However, it causes the problem in 
the later processing, such as triangulation, smoothing. In order 
to solve the problems previously listed, a spherical projection 
based on triangulation algorithm is developed to process the 
data separated by scanning station. After each station’s point 
clouds are processed, they’re merged together. 
1. Triangulation 
Because the points in a station can be seemed as one laser 
centre measured, around which a spherical space is given so 
that it can be consequently defined. Then the points are 
projected onto the sphere surface, and a closed triangulation 
based on the sphere surface is made. As the triangulation in 
sphere space is similar to 2-dimensional triangulation, the 
algorithm is easier and faster compared with the 3-dimensinal 
triangulation algorithm. After the spherical triangulation, the 
data are projected back to its real position with the triangulation 
structure reserved, based on which mesh object can be created 
easily.