Figure 1 “time 2" time 3" time 4^ time 5^ time Simulation 
data data data data data data 
Depth Accuracy Fig.10 | 0.33mm 0.41mm 0.46mm 0.39mm 0.48mm 0.74mm 
(Difference of K, from the 1* time) (0.0) (8.10e-5) (1.18e-4) (4.25e-5) (1.36e-4) (3.0e-4) 
Depth Accuracy Fig.11 0.38mm 0.52mm 0.46mm 0.39mm 0.41mm 0.68mm 
(Difference of K; from the ]* time) (0.0) (-9.79e-6) | (4.74e-6) (1.29e-6) | (-2.23e-6) (2.0e-5) 
Depth Accuracy Fig.12 | 0.38mm 0.41mm 0.41 mm 0.38mm 0.38mm 0.74mm 
(Difference of x, from the 1* time) (0.0) (12um) (-14um) (-7um) (-8um) (70um) 
Depth Accuracy Fig.13 0.38mm 0.38mm 0.43mm 0.41mm 0.41mm 0.79mm 
(Difference of y, from the 1* time) (0.0) (-6um) (21um) (-15um) (17um) (70um) 
  
  
Table3. The non-uniformity of accuracy by the reproduction 
And the Table 3 shows the accuracy of this reproducibility. In 
order to assess the accuracy of the Flat Sheet calibration we 
made 3D measurement of the control points of 3D Target-field 
by using the calibration parameters obtained by the Flat Sheet 
DC1000-processed calibration (f — 7.2mm, Camera length: 
1313mm, Base length: 528mm), where the accuracy is the 
lowest. As you notice, everything beautifully falls within the 
range of 0.5mm (about 1/3000). 
The data in the far right column is the result of the simulation 
case where the non-uniformity of data got worse more than two 
fold (Simulation data 7.2mm in the graph). But even in this 
simulation the accuracy was still as high as 0.7mm. 
4. 3D MEASUREMENT BY ON-SITE CALIBRATION 
Based on the results so far obtained, we went to the actual site 
of a stone piling wall and performed 3D measurement (Figure 
14). We tested at 3 points within the focal length range 
between 72.mm and 50mm: one at comparatively wide angle 
(9mm), the second near the middle (30mm) and the third at 
comparatively telephoto angle (42mm). The size of the 
photographing area was about 2mx2m. 
The photographing conditions are on the Table 4. 
  
Figurel4. Stone piling wall (Test field) 
The precision actually obtained is in the Table 5. 
We succeeded to obtain the satisfactory precision which is 
within 2mm for the targeted lcm. We can also note that the 
  
  
  
  
  
  
  
  
  
  
>>. Focal Camera Base accuracy was much higher (in our experiment nearly 10 times 
length | distance H | length B higher) than when we did not make calibration. 
Casel: wide (mm) 9 2586 774 
Case2: middle (mm) 30 6494 1904 
Case3: telephoto (mm) 42 7516 1706 oxy = H/f*op (3) 
ôz = H/f*H/B*6p (4) 
Table4. Conditioning for photographing and analyzing 
We made the test following the flow chart of Figure 12, using where &xy = resolution of one pixel on surface 
PI-2000 for 3D measurement. We also measured 49 points with &z = resolution of one pixel in depth 
our Total-Station (measuring accuracy: 4—1mm), to 8 points of H = camera distance, B = base length 
which we applied Self-calibrating bundle adjustment to f = focal length, Jp = pixel size 
calculate the focal length, while the rest of the points served to 
assess the accuracy. 
Table5. The result of accuracy test 
  
  
  
  
  
  
  
  
  
Focal Position Focal length (Analyzed | y-Parallax at relative Surface accuracy | Depth accuracy Óxy oz 
value) [mm] orientation [um] [mm] [mm] [mm] [mm] 
Casel: wide 8.83 1.5 0.7 1.4 1.0 3.3 
(with no correction) (8.8) (8.4) (13,5) 
Case2: middle 30.04 0.8 1.3 1.8 0,7 2.5 
(with no correction) (6.7) (9.6) (15.2) 
Case3: telephoto 43.21 1.3 1.4 1.9 0.6 2.6 
(with no correction) (6.8) (11.3) (18.8) 
  
  
  
  
(calibration data in the widest angle : 7.2mm) 
58.