A testfield with known control points is mounted below the 
production line. Between two measurement cycles the 
system can be oriented again. For the control points 
circular alluminum targets are used which are relatively 
robust against mechanical disturbances. Retro-reflective 
targets can not be used in this environment. 
2.3 Software Components 
The system is running under Microsoft NT 4.0. The 
control software has been developed under Borland 
Delphi 3.0 which is able to integrate C-written procedures 
and libraries such as standard software for image 
processing. 
3 SYSTEM CALIBRATION 
3.2 Camera Model 
Processing of analog video imagery yields to significant 
problems in image geometry. Line jitter and image affinity 
are well known (Beyer 1992). Standard TV lenses often 
do not meet photogrammetric requirements and have 
significant orders of radial and decentric distortion (Fryer 
1996). 
For this application the following set of parameters is 
used to describe camera geometry: 
image coordinates with respect to 
X'=Xp—X es ; 
the principle point 
V'=Yp-Y9 
X"=x+A 7 image coordinates corrected by 
p m MR frame grabber affinity whereby 
ye y Ava hen ie tg 
Ay aff =0 
The correction terms Ax' and Ay' are functions of 
different distortion effects. Radial-symmetric distortion is 
defined by 
" " AF n " " Ar? n 
Ax sym — X Ay sym — RL dren (2) 
r r 
whereby 
Ar P Arr, d) Aor (n 1M) A, r'(r* 6) 
sym 
and  r^image radius 
ro: Second zero-crossing 
Decentering distortion is given by 
Ax", = B, (rh 239 24, 2D Xy" (3) 
Ay" = B, Ar? +2 py"? )* 2B, x". y" 
asy 
The complete correction terms of image coordinates is 
then 
244 
Xx" Ax! AX! 
sym asy 4 
y2J" -Ay tA e 
This model is  well-proven for different kind of 
photogrammetric cameras such as analog film cameras 
or digital still-video cameras (Godding 1995). In the case 
of low-cost video cameras, analog signal transfer and 
separate video digitizing the model yields to the following 
problem. 
All parameters of interior orientation are defined with 
respect to the principle point, which should be the point of 
auto-collimation. This is true for most cameras where the 
the projection center is very close to the optical axis. In 
the case of large offsets in the principle point position the 
parameters of radial distortion 4j, A», and the parameters 
of affinity and sheering, C;, C» are highly correlated with 
the coordinates of the principle point x s, y 'o. 
In order to determine the parameters of interior 
orientation by self-calibrating bundle adjustment it is 
nessecary to introduce an a priori correction to the 
principle point coordinates. lteratively the measured 
image coordinates are corrected by the residuals of 
bundle adjustment until the whole system becomes 
stable. 
  
Figure 5: Distortion in original video image 
  
Figure 6: Video image corrected for distortion 
Figure 5 ant 
large radial 
after self-ca 
3.3 Testfie 
The CCD-c 
images of tf 
point and e 
7), the ob: 
corrected i 
interior orie 
adjustment 
system rea 
step of ce 
variations ii 
much pooré 
distortion. 
100 4 
50 4 
-50 4 
radial distortion dr’ [um] 
-100 4 
  
-150 - 
The me 
transforma 
4m) has I 
Sxy=+1.2m 
result is ec 
meets the | 
4.2 Point 
The measi 
accuracy f 
(e.g. Luhi 
application 
analysis al 
in the im 
positions 
where the 
of the e» 
accuracy t 
the pixel si 
4.3 Edge 
The actu 
measurem