ABSTRACT
decesive measure for selfdiagnosis.
1 PERFORMANCE FEATURES
The presented approach is a generic solution for the AIO.
Images from different types of cameras can be processed.
The following performance features are implemented:
e The procedure is fully automatic. Except for the
approximate image resolution and the camera type,
no additional information or approximate values are
needed.
e Gray level and color images can be processed.
e The orientation of the image is automatically recog-
nized. :
e Whether the image is positive or negative is automat-
ically recognized.
e Robust algorithms guarantee correct results even in low
contrast images.
e Reliable self-diagnosis enables automatic recognition of
unsolvable situations.
The fully automatic process of the reconstruction of the in-
terior orientation expects the following:
A The image in digital form, including
— an image pyramid and
— the approximate resolution of the image.
The pixel size is usually known from the scanning
process with an accuracy of +1pm which is more
than sufficient.
B The camera type and the usual camera calibration in-
formation, including
— Fiducial mark patterns and
— a pattern of an unsymmetric feature for the
recognition of the orientation.
Except for the calibration data all the information specific to
one camera type is stored in a so-called camera description
file, which is available for all the conventional camera types.
THE AUTOMATIC INTERIOR ORIENTATION AND ITS DAILY USE
Wolfgang Schickler
Zoltan Poth ?
!/ Analytical Surveys, Inc., Colorado Springs, Colorado, USA, wolfgangQanlt.com
2) Carl Zeiss, Oberkochen, Germany
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
1)
Commision lll, Working Group 3
KEY WORDS: Photogrammetry, Orientation, Automation, Softcopy, Matching, Digital
We present a fully automatic and operational procedure for the reconstruction of the interior orientation of digitized aerial
images. The main task of the Automatic Interior Orientation (AIO) is the robust localization of the fiducial marks in the
digital image and the reliable estimation of the transformation between image and pixel system. We paid particular attention
to robustness of the procedure. Results of extensive tests confirm the high reliabilty of the approach and also the internal
2 CONCEPTUAL ASPECTS
There are eight different possibilities (orientations) for how
the images could have been placed in the scanner for dig-
itizing: wrong reading or right reading with four different
90° rotation respectively. For a generic solution to recognize
the orientation of the image we expect a unique asymmetric
feature in the image. With a template and the coordinates
in plate system of this asymmetric feature, it is possible to
match the template with all eight possible positions where it
could appear depending on how the image was scanned. À
classification of all the matching results leads to the orienta-
tion of the image.
Figure 1: Schematic illustration of a digitized image
All conventional cameras have at least 4 symmetrically placed
fiducial marks, so that at a certain position a fiducial mark
can be found whose shape is known, independent of whether
the image is right reading or wrong reading and from 90°
rotations of the image. We call those orientation invariant.
The fiducial marks 1 to 4 in figure 1 fulfill this criterion, while
the shape of the fiducial marks in the corners in this case is
dependent on the orientation in the image.
Our approach is based on the location of at least four ori-
entation invariant fiducial marks in the image, without any
prior information. After this, the transformation between the
pixel and the plate system can aproximately be estimated,
except for an unkown factor of 90° rotations, and possibly a
746
to d
As f
and
ages
an a
cial
an L
grou
loca
ing |
addi
usin
On |
for t
Fig.
TOF
marl
level
lowe
cial
unde
imag
don'
left,
Figu
with
mark
The
