983
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B5. Beijing 2008
2. Smoothing
After the triangulation, smoothing can be easily applied to the
mesh objects. Two options can be used for smoothing, namely
feature preserved smoothing and common smoothing (Attali et
al., 2003).
3. Hole filling
Because of the shelter of other objects and the scanning angle
limit, there are lots of small holes in the mesh generated. The
goal to fill a hole is to remove it from the mesh.
4. Decimating
This step aims to reduce the number of points. Similar to the
smoothing, decimating often reduces the accuracy of shape
because it will remove some triangles in the mesh. Feature
preserving decimating is used, which can reduce the amount of
triangles with features of model well reserved.
5. Model merging
After filling a hole, the mesh object from difference scanning
station should be merged into a whole. So the data needs to be
transformed and put together. Sometime, there exist some small
holes among the meshes and those duplicated triangles. By
rewrapping the data with few editing can solve the problem.
2.4 Multi-baseline Rotating Photogrammetry
This method is introduced to acquire the image data of the nine
storied Mogao Cave, including the ground acquiring way and
elevator acquiring.
3. DATA PROCESSING
3.1 Characteristics of the Mogao Caves
The Mogao Caves owns its characteristics which cause plenty
of problems to our work. First of all, most of the immovable
caves own an irregular shape with a lot of tiny apertures and
sheltered places without visibility. And this particularity
directly makes it more difficult to acquire its data even by the
most flexible way of laser scanning. Thus much more work has
to be done but hardly anything fruit returned.
Secondly, after the acquiring work, the entire data amount arose
even more difficulties for processing. For instance, even for a
signal cave, with a scanning precision of 2mm, its whole data
amounts to some 5 millions triangles, and the texture
information is even more fruitful, which requires both a detailed
recording with high accuracy and much relational information
with geometric structure stored.
Thirdly, a distinctive difference with a common object lays in
its texture. All the fresco and painted statues on the wall can be
treated as a rare book recording all the thousands of years’
culture collision and variance, based on which we can pick up
all those beautiful stories plus the massive culture beyond. Thus,
an especial requirement on the detail information and accuracy
of the texture is rising up, which causes plenty of problems for
our acquiring and processing.
3.2 Modelling
In this workflow, 11 stations with 5m interval between each two
nearby are set at some 50m away from the nine-storied Mogao
Cave. Then each station takes a rotated photographing of the
building. Four ribbons are laid on the ground. While on the
elevator, the same method is implied using 5 stations with 10m
interval between. In order to gain its absolute size, plenty of
controlling points given by the control surveying are laid both
on the surface and around the building, which are used as
controlling data for later photogrammetry processing. Also, a
parallel photographing method is chosen to acquiring the image
of the outside wall, whose length measures to be almost 1.6 km.
Among the difficulties above, a lot of work has been done to
deal with them in the processing process. For instance, towards
the laser scanning data for modelling, a reasonable workflow to
its construction process has been set up after fully considering
the needs of the Mogao Caves. Briefly speaking, this process
includes several main steps as following. Firstly, laser scanner
has been used to acquire 3-D point cloud data by piece, then
these pieces of data is joined together according to some
position relationships.
Figure 4. The point cloud of Cave No. 158
Figure 3. The sketch of multi-baseline rotating photogrammetry
Figure 5. The grey model of Cave No. 158
