il 2004 
due to 
ICy in 
atively 
p of 1 
vn to a 
1e laser 
  
ages of 
imately 
/0 MB 
II, we 
mages, 
It took 
urch Il. 
cording 
due to 
mpling 
and the 
et was 
ng step 
r, but it 
enough 
can the 
another 
F4 FLE 
record 
provide 
36.86 x 
Around 
pictures 
ce them 
s walls, 
raphing 
cases, 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B5. Istanbul 2004 
  
photographs were taken from a slanted angle. All of the 
photographs were taken as single, not stereo images since we 
had just one set of the DCS Pro Back. 
4. DATA PROCESSING 
41 3D Models from Point Clouds 
To align and merge the multiple 3D images to produce 3D 
polygonal models, Polyworks from InnovMetric was used. First, 
the images were translated from a local coordinate system to a 
global coordinate system based on the coordinates of the 
reference. points included in the images using the function 
provided by LPMSCAN (software from Riegl) to control the 
scanner. The coordinates of the reference points were obtained 
from a conventional topographic survey. Secondly, the 
translated images were imported into Polyworks and then 
merged into one polygonal model. The size of the polygonal 
model of Church II in the POL format was about 2,560,000 
polygons. The file size was approximately 70 MB. This would 
be even larger if the model was saved in VRML format. 
Since the size of the point-cloud data for the 3D model of 
Church III was huge, we first partitioned the church into several 
sections, i.e., the apse, the south wall, and other areas, and then 
created a 3D model corresponding to each section. These 
separate models were then merged into one. Some section 
models were based on the original point-cloud data and others 
used the reduced data. The total size of the 3D polygonal model 
of Church III was approximately 6,300,000 polygons in the 
POL format. The file was about 174 MB in size. 
  
Figure 3. 3D model of Church II: close-up of apse 
Figures 3 and 4 show 3D models of Church II and Church III, 
respectively. Note that these models have no texture, but color 
information is attached to the vertices. This appearance is due 
fo vertex color, i.e. color captured by the LPM-25HA scanner, 
and considerably differs from the original color. 
403 
  
AE 
€" 
  
Figure 4. 3D model of Church III: bird's eye-view (above) and 
close-up of west wall (below) 
4.2 Point Cloud with Color Information from Photographs 
Usually, RGB images captured by laser scanners are not as high 
a quality as required for many applications, and even worse, 
some scanners cannot acquire color images. Therefore, other 
methods of enhancing representations of the appearance of a 
scanning target are needed. One popular technique is texture 
mapping and another is the use of point clouds with color per 
vertex. 
Point-based rendering and modelling are currently important 
research topics in computer graphics (Pauly et al. 2003; 
Zwicker et al. 2001; Rusinkiewicz et al. 2000). This is because 
laser scanning technology is making it easier to obtain dense 
point clouds. Hence, the size of the polygon models being 
output has become larger and larger. However, the tools and 
infrastructure for handling large 3D polygonal models are not 
yet sufficiently developed. Rendering and modeling using point 
clouds is attractive since it reduce the data size drastically 
compared to polygonal models and would therefore be suitable 
for use on the Internet. 
We tried to improve the quality of the color information 
included in the 3D images scanned by the LPM-25HA laser 
scanner with the high-resolution images taken by the DCS Pro 
Back. Each image was analyzed using a corresponding 3D 
image. The user specified more than six matching points in both 
a photograph and a 3D image so that the photograph was given