rfaces
stem
ordinates
dditional
its of the
ze in one
| geomet-
in optical
ear with
| the en-
dditional
ie. photo-
| is fixed
rs of the
ic image
ic optical
he image
ce to the
ns has to
he image
sical part
he trans-
re as fol-
ric optical
'ammetric
n;
netric sys-
pil;
iple point
ollimation
All data to solve this three dimensional vector equa-
tion can be taken from the reports describing the opti-
cal design and the camera calibration, respectively.
The equation is identical for the transformation of the
parameters of additional refracting surfaces, if these
are defined in a vector oriented form.
3.1.4. Computation of Target Image Models and
Definition of the Photogrammetric Image Points
Two conditions must be fulfilled by the mathematical
models of the target images: they must allow the
identification of the intersection points of the corre-
sponding rays and they have to be realistic models of
the density distribution in the real target image.
Using ray tracing for the computation of the target
image models fulfills the first condition: Starting at
an object target a representative number of rays is
traced numerically through the optical system and in-
tersected with the image plane (see in detail Kotowski
1988 pp.324ff and Pegis 1961, pp.8f). The type of each
ray is defined in advance. This means, each of the
traced ray is either a principle, a meridional, a sagit-
tal or a skew ray, it starts with a certain incident
angle from the object target and has one specific
wavelength. One of these rays, the principle ray of
one colour, is defined as the corresponding ray be-
tween object point and image point. The intersection
of this ray as well as of all other rays results in a two
dimensional distribution of intersection points, the
first state of the target image model. The intersection
point of the corresponding ray, well defined in this
model, is defined as the photogrammetric image
point.
If the density distribution of all intersection points in
a target image model is equivalent to the density dis-
tribution of the real target image, the second condi-
tion is fulfilled. To achieve this representative
distribution, the correct selection of rays to be traced
is essential. Their number and their spatial distribu-
tion have to approximate the radiation that is sent
back from the object target and darks the sensor sur-
face sufficiently.
The abstraction from the radiation to be modelled to
the distribution of discrete rays is processed in three
steps. In the first step only that cone of radiation is
modelled, that is assigned to only one wavelength
and starts only from the center of the object target.
The resulting intersection points in the image plane
are equivalent to the point spread function, proposed
in this shape by Herzberger (Herzberger 1947, 1954,
1957), called a spotdiagram. To achieve a realistic
distribution of the rays one of the pupils is divided
into surface elements of equal size and each ray is as-
signed to one of these pupil elements (Herzberger
1958, p.106 und Cox 1964 p.378).
In the second step the object target itself is divided
into small surface elements, which are of equal size,
if a constant radiation can be assumed over the whole
object target. Starting at the center of each of these
surface elements, one spotdiagram is computed, lead-
ing to a group of overlapping spotdiagrams in the
image plane. This distribution of intersection points,
accumulated for one wavelength, shall be called the
monochromatic target image model.
In the third step the whole spectrum of wavelengthes
passing the optical system is divided into constant in-
tervals. For each of the resulting wavelengthes the
monochromatic target image model is computed as
described in the previous paragraph. The spectral
characteristics of the illumination, the target radi-
ation, the transparency of the optical system and the
sensor surface have to be considered by introducing
an adequate weight for the rays of each of the several
wavelengthes.
So, the required ray tracing for the computation of a
target image model is complete. The three accumu-
lated varieties of intersection points are overlayed by
a raster with a raster field size equivalent to the pixel
size of the real digitized target image. The intersec-
tion points are counted in each field separate and the
number of points in one field is interpreted as a den-
sity value, resulting in a digital image with a certain
density characteristic.
To represent the density characteristic within the tar-
get image realistically, the number of traced rays
must be large enough. To prove this the difference be-
tween the density values derived from two intensifi-
cations of numbers of traced rays is computed: If this
difference is still significant, a further intensification
has to be performed, i.e. more rays have to be traced,
as long as the density characteristic does not change
any more.
As mentioned above, for each target the principle ray
of a constant wavelength, starting from the object tar-
get center, is defined as the corresponding ray and its
intersection point with the image plane is the photo-
grammetric image point. Its position within the digi-
tal target image model is known; to identify it in the
real target image, it has to be transferred there.
3.2.Identification of Comparator Coordinates
for the Defined Image Points
To be able to identify the photogrammetric image
points in the real target image, this and the digital
target image model have to be matched, e.g. by least