5. Istanbul 2004 
>ublishing house 
"ublishing house 
) Reconstruction 
r’s dissertation. 
Information of 
metry. 
Complex Shapes 
ssing and 3D 
'"hotogrammetry, 
Information 
ree-Dimensional 
| Trunk from 2D 
on and Visual 
nage Sequences. 
Multi Images. 
| Image Sensing 
hotogrammetry, 
Is with Desktop 
Electronics, Vol 
ek, L.G. Using 
es of the Human 
tion on Pattern 
lo.11.November 
on from Image 
. Lecture Notes 
2002: 50-57. 
D Point Clouds 
erPoint of Oral 
eminar on New 
y. September 
Reconstruction 
es. Proceedings 
Data Processing 
2002. 
CAMERA CALIBRATION TECHNIQUE BY PAN-CLOSEUP EXPOSURES FOR 
INDUSTRIAL VISION METROLOGY 
Harutaka Imoto *, Susumu Hattori °, Keiichi Akimoto ‘, Yuzo Ohnishi * 
* Dept. of Production Technology Development, Ishikawajima-Harima Heavy Industries Co.,Ltd., Yokohama, Japan 
? Dept. of Computer Science, Fukuyama University, Fukuyama, Japan 
* Dept. of Control Engineering, Shikoku Polytechnic-College, Marugame, Japan 
? School of Urban and Environment Engineering, Kyoto University, Kyoto, Japan 
Working Group V/1 
KEY WORDS: Industry, Photogrammetry, Calibration, Bundle, Camera, Distortion, Targets, Close Range 
ABSTRACT: 
A high precision and easy-to-use CCD camera calibration technique for industrial vision metrology is discussed. A well-known 
method is self-calibration by convergent camera configuration of a two- or three-dimensional target field. Only with this technique 
the central part of a sensor area is precisely calibrated, but off the centre the precision rapidly deteriorates. The presented technique 
is a simultaneous adjustment of both pan and close exposures, which compensates the lack of distortion data in the fringe area of the 
sensor and offers both uniform and high-precision calibration. Some patterns of camera configuration are compared in an experiment 
in terms of the precision and its uniformity over the sensor. And the combination of convergent pan exposures and vertical close 
exposures is proved the best. 
1. INTRODUCTION 
In industrial vision metrology with a single camera, high 
precision can be obtained by self-calibration, if a measurement 
configuration is good or in other words a measurement network 
is strong. But in many situations possible camera configuration 
is limited, targets are often not well distributed in space (even 
after supplement targets are added), and exposures might be 
reduced in number to save processing time. 
If a network is weak, pre-calibrated interior orientation 
parameters are necessary, which are incorporated in bundle 
adjustment as weighted observations. Especially in the case of 
off-the-shelf cameras, the body is a bit fragile and therefore 
frequent camera calibration is required, and in practicability a 
cheap, time-effective and high precision technique is 
indispensable. 
A conventional and reliable calibration method is a bundle 
adjustment of images of multi exposures over a field of 2D- or 
3D control points (Hattori, 1995). Self-calibration of images 
taken in convergent camera configuration has been reported to 
be à good substitute in the case of no control points. But only 
with this technique, though the central arca of a sensor is very 
well calibrated, the precision of parameters in the fringe of the 
sensor is deteriorated, since smaller number of common targets 
is captured in fringe areas. This causes the precision shortage 
not only industrial applications, but in conventional sterco 
measurement, where the entire sensor area is equally used. 
This paper presents a self-calibration technique of simultaneous 
adjustment of images taken in different exposure distances. The 
basic idea is as follows: The object space coordinates can be 
measured from images of convergent exposures over the target 
field at remote stations. Images capture the field at the sensor 
centre. Then by use of these object coordinates the distortion 
functions can be precisely evaluated from images taken at close 
stations. The target images are uniformly distributed over the 
sensor area. These two pan and close sets of images are 
simultaneously adjusted. Thus it is expected that the same 
effect as the calibration using a 3D control field is obtained and 
distortions are uniformly compensated up to the fringe of the 
sensor. 
In the following section, some combinations of camera 
configurations are compared by an experiment. As the result the 
self-calibration of images taken in a combined configuration of 
convergent pan exposures plus vertical close exposures shows 
the best precision. 
2. THE PROPOSED CAMERA CALIBRATION 
METHOD 
2.1 The distortion model 
As a model of lens distortions, well-known Brown parameters 
(Brown, 1966) are used. 
Ax z-X, Re +K,r*+K,r® X--x,) 
ep fe? + 2lx- x, y be 2P, (x=, Y» - »,) (1) 
Ay=-y, + (Kır? +K,r'+K,r" X -y,) 
+2P (x Fu Y» = )+ P, v + 2(y EY y } 
where (x, y) are image coordinates of an object point, (x,, y,) are 
coordinates of the principal point, and r^ (x-xJ t (y vp. K,, K; 
and K; are coefficients of radial distortions and P, and P; are 
those of tangential distortions. It is assumed that the principal 
point oma with the centre of lens distortion.