International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B5. Istanbul 2004 
  
  
X,y,z coordinates for the reference points. Matching and 
analyzing were done using software we developed. Figure 5 
shows the original RGB color of a 3D image and Figure 6 
shows the partially improved 3D image. As a comparison shows, 
the results were reasonably good. 
Although the result was quite good, it is important to note that 
there is not just one photograph corresponding to a 3D image 
since a laser scanner can scan a very wide area at once. For 
example, the LPM-25HA laser scanner can cover 360° 
horizontally. Color adjustment between several photographs is 
thus required. If the target is an object that can be placed in a 
room or the interior of a historical building, it is easier to 
control the illumination during shooting. However, if the target 
is located outside, as in our case, it is difficult to control the 
lighting conditions and the laborious task of adjusting the color 
is required. 
Modelir 
  
Figure 5. Original RGB information included in the 3D laser 
data 
  
Figure 6. Improved RGB information (top-right side) 
4.3 3D Models with Texture Mapping 
To increase the quality of their appearance, we added texture to 
the polygonal models. The texture images were created from 
high-resolution image data taken by the DCS Pro Back. For 
texture mapping, PI-3000, a digital photogrammetry software 
from Topcon (kochi et al. 2003), was used due to its 
outstanding features such as follows: 
|. It can process pictures taken by a standard digital 
camera. 
2. It runs on a note PC 
3. It can calculate Digital Surface Models (DSM's) fully 
automatically ! 
In addition, it does not require a fixed base or other set up 
conditions if users do not need sub millimeter accuracies. This 
was very important for our case since the pictures were not 
taken by a specific setup condition prepared beforehand. 
As there was a huge number of points, reduced point-cloud data 
for Church IT in ASCII text format was imported into PI-3000. 
A photograph was chosen and also imported into PI-3000 and 
analyzed using bundle adjustment. To align the 3D polygon 
model and texture images, three reference points, with global 
coordinates obtained from a topographical survey, were used. 
Figure 7 shows the textured 3D model of Church lI. To indicate 
the completeness of the matching, the point cloud is shown on 
the right. 
  
wp KL Eee Auc AXI rA 
DUE dd 3 Aci 
¥ i 
PUISE ENS SUES 
Figure 7. Textured 3D model of Church II 
4.4 3D Models from Digital Photographs 
We attempted to create 3D models from digital photographs 
using PI-3000. To model the semi-dome and apse of Church II 
three-dimensionally, two pictures were selected. These pictures 
were taken at a distance of 16 m and the baseline length 
between them was 6 m. The focal length was 52 mm. The 
theoretical resolution was therefore 3 mm for XY and 8 mm for 
Z. Note that these images were not taken by a specialist in 
photogrammetry. They were more like tourist photographs 
except that a high-end camera was used. 
A TIN model has been automatically created and the number of 
produced points was about 260,000. The resolution is 2 mm. 
The result is shown in Figure 8. As can be seen, the small 
details are well modelled. There is a crack on the wall (see 
Figure 8 above) that is 1-cm wide for the most part. The actual 
resolution in this case is thought to be less than 5 mm. Figure 9 
illustrates different views of the same model. The shape of the 
semi-dome is well modelled, although it is difficult to model 
shapes such as a dome, as in this case, using conventional 
photogrammetry. 
  
[Internati 
LEE 
    
  
  
Figure € 
X 
  
Figure 
To eva 
elevatic 
Byzanti 
measur 
string, : 
In cont 
photogr 
analyze 
process 
digital 
conditi 
with a