SRA SE VLE aT 
  
BR Toi mas Master on 
  
aaa PA 
T meh um = TUE c 
m" 
: Genf PET 
per T | 
EAE GEE hd | 
  
var 
007523638 
| [ose EI | 
  
| yes pues 7 
M yam | MONEMIQGI OOUNE — a... A UND | Mott panem 
OENOSAORAE — p - GEESOSADRAS à | BRR ps dai | 
| pone lumi poer m | puer lum] be Be | Li B cd | 
  
Figure 6. Lens distortion 
The Figure 6 shows the lens distortion and the interior 
orientation parameters when the Dimage7's angle is the widest 
(focal length 7.2mm: equivalent to 28mm in 35mm format). 
2.2 Precision Result of Flat Sheet Calibration 
First we obtained the camera calibration parameter by DC-1000 
and then made 3D measurement of the control points of 3D 
Target-field (Figure 1) to assess the accuracy or precision. 
The index is the camera distance divided by depth precision, as 
the standard deviation of residual of control points. 
Since the lens-distortion is greater and the precision is smaller 
in the wide angle than in telephoto, we experimented with the 
widest possible angle (7.2mm). 
As to this index, it is sufficient to make it 3000 in order to keep 
the targeted accuracy within lcm in the actual site. 
The accuracy of measurement as influenced by the camera 
angle of incidence is shown in Figure 7. 
As to this index, we found that in order to obtain the accuracy 
of 3000 we simply have to limit the camera angle within 10~30 
degree. That is why in this experimentation we first tentatively 
and practically set the photographing angle to approximately 15 
degree. 
Distance/Depth eu» Targeted Accuracye G)00 
3500 
3000 
2500 
2000 
1500 E- 
1000 
500 
0 
  
5 10 15 20 25 30 35 
Angle [?] 
Figure7. Accuracy of measurement as influenced by 
the camera angle of incidence 
  
  
epu test A mizio» testB TargetedAccuracy 
5000 prssmm————————————————————————— 
4500 [rmbt nlm eem 
4000 e 
3500 
3000 
2500 - 
2000 —— . oiii cote 
1500 E———— ———————— * E 
1000 —— : ———————— - 
500 }— : voe 
t TT es 
0 2 5 8 10 15 
Warp [mm] 
Figure8. The influence of the warp to the accuracy 
Moreover, Figure 8 shows the variance caused by the warp of 
the flat sheet. As seen from the Figure 8, the warp up to 5mm 
can keep the accuracy within 3000. This is sufficient value for 
justifying the use of the sheet for all practical purposes. 
Such variance would never occur, if we make calibration on the 
target field in the LCD monitor. 
Again the Table 2 shows the difference of the accuracy as well 
as the time spent between the calibration made on the flat sheet 
and the calibration made on 3D Target-field. 
The accuracy on the flat sheet is a little inferior to that on the 
3D Target-field, but the time required from taking 5 pictures up 
to the completion of the analysis is only 10 minutes. 
  
Target Used Depth Accuracy Required Work-Time 
/Camera Distance | including photographing 
  
Flat Sheet 1/3000 ~ 1/5500 About 10 minutes 
  
  
  
3Dtarget-field | 1/4000 ~ 1/11000 About 2 hours 
  
  
  
Table2. Accuracy and work-time 
3. ON-SITE CALIBRATION 
On the actual site, if we can not use the zoom and AF functions, 
we can not obtain satisfactory image quality and accuracy. To 
solve this problem we developed the on-site calibration system 
and tested its efficiency. 
3.1 Method of On-Site Calibration 
We first start with the preparation in the office. 
It is to calculate beforehand the calibration parameters at the 
several focal positions with the flat sheet. Then interpolating all 
these parameters with approximate curb by the least squares 
method, we obtain the parameters of the curb line. 
And then on the site we take the pictures and process their data 
through the Self-calibrating bundle adjustment to calculate the 
focal length of the image. 
Finally, out of the focal length thus obtained we calculate each 
of the parameters and with these calculated values as the 
interior orientation parameters of the camera we can make 3D 
measurement. 
The flow of the work is shown on Figure 9. 
-56— 
jd be eh. died A^ m 
Pd ped ped i