e 9 GCP are given (Case 4): Camera calibrations were 
performed by the bundle adjustment using 9 ground control 
points. 
Table 2 shows the R.M.S.E. for XY and Z coordinate for each 
case. Small differences can be found between the casel and 
case2. Nevertheless, it can be said that the case 1 is useful and 
convenient in severe condition that can't touch on object. If it is 
possible to measure only a distance in object field, the case 2 is 
recommended for obtaining high accuracy. The case 3 shows 
almost the same accuracy with the case 4 even given two height 
points. Then, it is concluded from the comparison with the 
results using 9 control points that the low-cost photogrammetric 
system using 3 million amateur camera is convenient method 
for digital photogrammetry. 
Table 2: Accuracy of the low-cost system 
  
  
  
  
  
  
R.M.S.E 
oxy (mm) oz (mm) 
Case 1 2.640 3.540 
Case 2 0.803 3.269 
Case 3 0.663 2.739 
Case 4 0.418 
2733 
0416 2.674 
  
  
  
Upper column in the case 4 is the result for the stereo images of 
the Casel and 2. Lower column is stereo image of the Case3. 
4. 3D MODELING 
4.1 Data acquisition 
3D modeling for the "Koma house" was performed in this paper 
as application of the low-cost photogrammetric system. The 
Koma house was built in 17th century (300 years ago), and 
designated as national important cultural assets in 1971. Stereo 
images for the Koma house were taken using the system with 
wide lens, and the detail procedure for 3D modeling are 
follows: 
+ Camera calibration was performed previously using test 
model. 
+ Stereo images for the Koma house were taken, and altitude 
for each image was measured by laser range finder. 
+ Lens distortion for stereo image was corrected using interior 
orientation parameters which were obtained previously. 
+ Feature points for relative orientation and 3D modeling were 
selected by manually. 
+ Stereo matching was performed by semi-automatically using 
SSDA (Sequential Similarity Detection Algorithm). 
+ Relative orientation was performed by coplanarity condition. 
+ Absolute orientation was performed using altitude. 
4.2 Texture mapping 
Figure 7 shows reconstructed wire frame model and Figure 8 
shows the 3D texture model. Following are detail procedures for 
texture mapping: 
+ Generating polygons using vortexes on the wire frame 
model. 
+ Calculation of image coordinates of vortexes for a polygon 
using orientation parameters. 
+ Cut textures for polygons from image. 
+ Textures are mapped to polygons on the wire frame model 
automatically. 
S. CONCLUSION 
Performance evaluation of 3 million amateur cameras for digital 
photogrammetry were investigated. At first, lens distortions for 
amateur cameras were investigated, and lens distortion can be 
corrected by the radial polynomial 3rd degree model. As a next 
step, accuracy aspects were investigated, and it can be seen that 
accuracy for amateur cameras are improved with the increase of 
resolution. 
  
Figure 7: Wire frame model 
  
Figure 8: 3D texture model 
However, the accuracy of more than 3 million digital cameras 
were hardly varied, and the accuracy of amateur cameras and 
professional digital cameras were almost the same value. Image 
quality was also investigated, and the obvious difference of 
image quality between 0.31M and 3.34M pixels can be found. 
On the contrary, it can't be found significant difference between 
3.34 and 6.10M pixels. Furthermore, the low-cost 
photogrammeric system which was developed by the authors 
were estimated, and sufficient accuracy can be obtained with 
only altitude are given. Finally, 3D modeling of important 
cultural assets using the low-cost system with 3 million digital 
camera was demonstrated in this paper as application of low- 
cost system, and it is expected that 3 million amateur cameras 
will supply the place of a professional digital cameras, and low- 
cost photogrammetric system will contribute to the development 
of the market in digital photogrammetric fields. 
-340-