The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Voi. XXXVII. Part B5. Beijing 2008
we discuss some of the strategies we developed for dealing with
the investigation on diseases of murals. Figure 1 is the pipeline
of investigation on mural diseases.
2. ACQUIRING LASER DATA AND IMAGE DATA
Since almost entire walls of the prayer-wheel gallery are
covered with murals, if only scanning single mural, we cannot
know it the position in the prayer-wheel gallery; if fine
scanning the whole prayer-wheel gallery, we can hardly the
huge scanned data. We employ a data acquisition planning at
multi-scale, using 3D laser time-of-flight scanner to capture 3D
data of the basic structures of the prayer-wheel gallery, and
using 3D laser triangulation scanner to fine scan the seriously
damaged murals. All laser scanned data was stored as range
image.
During data acquiring, view planning is one of important tasks,
and it plans the viewpoint locations to determine where to take
each individual scan. This requires choosing efficient views that
will cover the entire surface area of the structure without
occlusions from other objects and without self occlusions from
the target structure itself.[Paul S.,2007].For the prayer-wheel
gallery, considering its long and narrow structure, the 3D laser
time-of-flight scanner stood in the middle of the prayer-wheel
gallery, and 360 degree scanned, and we captured the laser
scanned data of both sides of the prayer-wheel gallery. The 3D
laser time-of-flight scanner was kept horizontal to ground to
scan the roof of the prayer-wheel gallery. We decided to overlap
adjacent scans from 30% to 40% - enough to align them in
software. We could compute the distance between viewpoints
based on the above demands. Figure 2 shows a floor level slice
of the resulting interior model of the prayer-wheel gallery. In
this experiment, scans were merged using reflective targets
scattered throughout the environment for registration.
precision to the detailed scanned models. We determined the
viewing locations as follow:
(1) Determine the region of the whole mural with a ruler.
(2) Compute the coverage area of each range image, while the
scanner is about 0.5m from the mural.
(3) To align range images in software, while selecting next best
view location, an overlap with known regions of at least 30%
was kept. Thus, we could plan views that the whole mural
needed. Figure 3 shows the data model of a mural using
Polyworks to align range images.
Figure 3. The data model of a mural using Poly works
to align 208 range images
On the other hand, the high resolution colour images contains
the rich surface details of the murals, they are the important
data source of mural disease analysis. Unhappily, colour images
only include 2D information, so generating the measurable 3D
orthophotoes are very important for mural disease investigation,
protection and restoration. We use a high resolution digital
camera with a wide-angle lens to capture the colour images of
the murals, and use parallellight to compensate for ambient
lighting. While capturing images, we tried our best to ensure
that either the horizontal or the vertical direction was taken as a
whole, only move the camera along the horizontal or the
vertical direction, and form a sequence of images with 60%-
80% overlap between two adjacent images.
COMPUTER-AIDED MODELING
After range scans and colour images were acquired, it entered a
lengthy post-processing pipeline, whose goal was to produce a
measurable 3D orthophoto.
(1) Blocked registration. One significant challenge we faced
in this project was the size of our datasets, for example, the
point cloud of the mural of figure 3 included 208 range images,
each range image contained 305269 points.We addressed this
problem by using a blocked method proposed by us. Firstly, all
the range images were blocked into different groups, and each
group included many range images. The range images were
registered in a group, and then the groups were aligned into a
common coordinate system.
