As a matter of course, calibration accuracy depends on 
pointing accuracy of image coordinates. Furthermore, 
most importantly automated targeting is required for real- 
time photogrammetry. Therefore, the following image 
processing procedure for this purpose was developed . 
Though a pointing accuracy less than 0.1 pixel is 
expected under good conditions ( low noise and target 
size)(Schaefer and Murai,1988), 0.3 pixel (=0.003mm) 
was adopted as Jp in this paper. The basic steps of this 
image processing procedure are shown in figure 3. 
Original Image 
" Enhancement " 
- smoothing - 
iti 
Cleaned Image 
" Segmentation " 
- optimal thresholding - 
e 
Binary Image 
" Extraction " 
- region boundary - 
ti 
  
Boundary Image 
" Recognition " 
- curvature of boundary - 
- area gravity of ellipse - 
- ellipse fitting with interpolation - 
  
Coord. List 
in 
Figure 3 Flow of automated targetting 
It may be seen from the results of this experiment that, 
1) with regard to 2D, accuracy of 1/1000 for camera 
altitude were obtained in case 1. 
2) accuracy for height improved 3096 in case 2. 
3) accuracy for 2D and height improved 8096 and 5096 
respectively in case 3. 
From these results, it is confirmed that the camera 
calibration method proposed in this paper is effective in 
reducing the geodetic surveying for control. 
As noticed above this method depends strongly on the 
verticality of a camera. For improvement on this point, 
the following are considered: 
  
1. Height for control points are given. 
2. Beginning with two distances and the parallel type in 
form, the following three additional types are adopted 
:Box , Cross and F types (Figure 4). 
The box type means utilization of 4 distances.The cross 
type adds one control point compared with the box 
type.Thus two more collinearity condition equations can 
be obtained. The F type is developed from the cross type. 
The remarkable difference in this improvement compared 
with case 2) in the first experiment is that the initial 3D 
ground coordinate for control points can be calculated 
using distance and height because control points are 
linked to each other. 
A second experiment was performed under nearly the 
same conditions as in the first case. Unknown 
parameters were computed from the following equation, 
loa) A [P.(40?)| + le (atta) 
+ EE ATZE DC: + wi] 
G, = 
  
wi - (3) 
Calibration became quite stable since it was possible to 
get high quality initial approximations. Consequently, the 
weights adopted in equation (3) were of equal value(=1). 
Table 3 shows the results for the developed types. 
Table 3 R M S E for developed types 
  
  
  
  
  
X,Y Z 
Box type +0,28 MM! +305 MM 
Cross type + 0.33 +277 
F type + 0.42 + 4.85 
equation(2) + 0.26 +2.63 
  
  
Accuracy, except the F type, nearly coincide with the 
values which were calculated from equation 2). 
Furthermore, considering the field condition, the cross 
type has the advantage of setting the control points. 
Then, in order to reduce any geodetic surveying in site, 5 
control points with a given height and 4 pipes with a 
known length were previously prepared (Figure 5). 
  
  
1 
© 
Ke 
e etre 
D 
poro OO © 
olo olod 
CIC COND 
OO OO 
Te 
O 
FR guwemc 
  
  
  
  
  
  
  
Ooln eoo 
OOO 
© Oo ploo 
error rod 
Qoo 
O 00 
D OO 400 
A B 
  
  
  
  
  
  
  
  
  
(b Cross type 
(c) F type 
Figure 4 Types developed from parallel type 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B5. Vienna 1996 
  
  
4. DIG 
In orde 
camera 
lens (f : 
with a ti 
  
  
Fig 
  
XC-75 | 
  
  
CD-5C