DETERMINATION OF THE POSITION OF CROSSES WITH THE SUBPIXEL
ACCURACY ON THE IMAGE TAKEN WITH THE CCD CAMERA.
by Józef Jachimski and Wojciech Trocha .
The University of Mining and Metallurgy (AGH)
Kraków Poland
ABSTRACT: Digital images of 512x512 pixels, containing the record of a flat test of the 9 crosses of high
contrast were registered using CCD camera model MTV 1081 CB. Centers of all crosses were computed using
different digital evaluation methods. The coordinates of digital image compared with the object coordinates
gave the standard error smaller then 0.1 of pixel. The description of experiment is given in the paper.
KEYWORDS: Camera CCD, Accuracy, Image Processing, Digital System, Close-range.
1. INTRODUCTION.
The nature of digital images stimulates the auto-
matisation of interpretation and geometrisation. Spe-
cially automatic reading of images of targets occurs
very promissing. There are already very good results
of automatic derivations of coordinates published
by Luhmann and Wester - Ebbinghaus [1986] or by
Cruen & Beyer [1987] or by Streilein & Beyer
[1991]. Being involved in the analytical plotters con-
struction (this using the anologue continuous tone
pictures and that using digital pictures displayed on
the monitor screen [Jachimski & Zielinski 1992]),
and having constructed the 6X6 cm reseau camera
[Jachimski & Boron 1992], it was quite obvious to
begin the research connected with the computer
supported reading of coordinates of targeted points
from the digital images. To begin we have taken
probably the easiest to solve problem - the automatic
reading of coordinates of the reseau crosses from
the analogue continuous tone pictures with the use
of CCD camera and analytical plotter.
The result of the first step of corresponding research
are presented below. The flat high contrast test of
9 thick crosses was recorded with the use of ordinary
CCD camera. Various approaches to automatic in-
terpretation and measurement of the position of the
images of target-crosses gave quite promising results
of subpixel accuracy better than 0.1 pixel.
2. THE DIGITAL CCD IMAGE OF THE TEST.
The digital image recorded with the use of CCD -
Mintron MTV - 1801 C camera is composed of 512
lines comprising 512 pixels each. The elements of
image matrix are written to the computer memory
as a monodimensional vector of 512X512 = 262
144 elements. Each pixel uses one byte of memory
to record the level of luminosity within the scale of
256 levels. The CCD image sensor of 6X4 mm size
produces rectangular pixels 12X8 um. The electronic
391
circuit in the camera collects the image produced
by the CCD sensors as a chain of electric charges,
transform it to the analog TV signal and transfers
it to the frame graber card in the slave computer.
The frame graber card transforms that TV image
again to the discrete form, and assigns one of the
0-255 numerical values to each electric charge
proportionally to its intensity. During above transfers
the errors of the camera and frame grabber clock
synchronization can occur. This errors can produce
same pixel lines shifts, resulting in assigning of wrong
values of intensity to certain pixel positions. There-
fore the digital images registered in the computer
can slightly differ from the images produced by the
CCD camera. The stability of images, which means
the repeatability of recording of several subsequently
one after another registered images of stable test
can depend on warming-time of camera. The ex-
periments have shown that relatively stable records
can be expected not before two hours from the
moment of camera work initiation (plug in).
To exercise the CCD image geometry a flat test of
9 crosses was produced. To the flat glass the 5 mm
wide and 40 mm long nontransparent bars were
glued to form 9 crosses. The crosses form 3x3
target test field of size 23x15 cm which fits well
the 512x512 sensors portion of the CCD camera
field of view. The test was measured using the
Stecometer from Zeiss - Jena. The edges of each
arm of each cross were measured in 6 section
perpendicular to the arm axis. From the Stecometer
readings the mathematical lines corresponding to
the arms axes were calculated, and coordinates of
the cross centers were found as the cross-section
of the pertinent mathematical lines.
Several times in various moments the test was
recorded using our CCD camera situated in the
position approximately perpendicular to the test
surface. The images were analyzed using various
analysis algorithms, and the image coordinates of
the centers of crosses were calculated. To compare
i
{
3
S
t
:
B
Y
1
d
|
a
S
:
v
b
io
i
