- Redundancy:
It is not known how much intelligence the human
operator uses when measuring the coordinates of a
specific point, but he or she uses certainly more than
any available matching algorithm. Thus, the blunder
rate for individually matched primitives can be rather
high. Efficient blunder detection is only possible if a
large redundancy exists in the system. In a recent
discussion on matching Ackermann (1996) coined the
phrase "replace intelligence by redundancy’.
- Integration of geometry and radiometry:
Since in digital photogrammetry the original observa-
tions are radiometric grey values, and the geometric
concepts of analytical photogrammetry are still valid,
an integration between the two fields is mandatory for
a good matching algorithm.
- Self diagnosis of each module:
Each of the modules within a matching algorithm
produces intermediate results which should be
checked separately in order to avoid an accumulation
of errors. In particular checks should be of both,
geometric and radiometric nature. Examples include
the epipolar constraint and the cross correlation coef-
ficient.
After this short excursion to the principles of matching
we will return to the main subject of this paper, namely
the automatic orientation of photogrammetric imagery.
In the next chapter, the interior and the exterior orienta-
tion, the latter split up into relative and absolute orienta-
tion, are discussed in detail.
3 AUTOMATIC INTERIOR ORIENTATION
In this chapter we are interested in reestablishing the
relationship between the pixel and the image coordinate
system. This task is usually referred to as ’interior orien-
tation’. Note, that this term is somewhat misleading, since
the actual parameters of interior orientation, namely the
calibrated focal length, the image coordinates of the
principal point and the lens distortion parameters are not
to be determined. They are the result of the camera
calibration procedure which is carried out for each ca-
mera prior to image acquisition. Instead, we are concer-
ned with the determination of a set of (usually six affine)
parameters for the transformation from pixel to image
coordinates.
300
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
For digital cameras the relationship between pixel and
image coordinates is constant and is determined during
the calibration procedure in addition to the parameters
of interior orientation of the camera. Only if film images
are scanned in a separate step (which is the case in aerial
photogrammetry today) the sought relationship must be
(re-)established for each digital image individually.
The pixel coordinate system of the digital image is expli-
citly given through the matrix of grey values. The image
coordinate system, however, is only implicitly given via
the fiducial marks (in some cameras réseau crosses are
available instead of fiducials). Therefore, the trans-
formation between pixel and image coordinates can only
be accomplished via the fiducials as identical points. The
automatic reestablishment of the interior orientation is
thus a semantic pattern recognition problem: one has to
find the centre of the pattern representing the fiducials
and ascribe each found pattern the correct fiducial num-
ber. Fiducials of different cameras are depicted in fig-
ure 2. Since the approximate location of the fiducials in
the image, their size, shape and brightness distribution
are known it is possible to design an operational, autono-
mous interior orientation procedure. Besides being fast,
accurate, robust, and reliable (see chapter 1) it should
work with (see also Schickler 1995b)
- imagery from different cameras,
- imagery from different scanners,
Figure 2: Fiducials from different aerial cameras:
top left: RMK-A, top right: RMK TOP,
bottom left: LMK, bottom right: RC 30
——ÓA-————
- ima,
- blac
- posi
- à ve
-.4ma
The la
during
placed
rent pc
found
as the
The ir
image,
ra typ:
distort
and th
digital
of eac
format
curacy
fiduci:
dancy
care n
blunde
In ord
steps |
- apf
« Sub
- pos
In cor
chapte
be dei
highei
the w
the re
neous
lends
resolu
and sı
Subpi
ject f
1984;
1994)
