anbul 2004
ent. It had
(d to know
1e moment
ition were:
at arose in
It.
ce] axes of
mputed, a
the force
' unknown
formation
thematical
f software
n suitable.
le hybrid
bservation
using the
h is based
(T (Kager
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B5. Istanbul 2004
2.1 Data Preparation
The small-size negatives (24mmx36mm) were digitised in color
at Sum with the VEXCEL ULTRASCAN 5000 scanner. The
digital images were imported into ORPHEUS and image
pyramids were computed for faster visualization purposes.
When dealing with non-metric images there are no fiducial
marks available that can be measured in order to create the
relation between camera space and image space. Therefore the
four corners of each image had to be digitised carefully and
would serve as fiducials. Unfortunately, in most of the images
the texture and contrast was very poor at the comer regions,
thus the corner fiducials were measured indirectly. A thorough
description of this procedure is given in Waldhauesl and Kager
(1984).
A three-parameter-model was employed for the transformation
between camera space and image space since fictitious fiducials
were used.
At this point the images had been assigned the same interior
orientation (assuming no refocusing of the camera), which still
was unknown. So, the next step comprised the initialisation of
the focal length. It was set to 43mm (= diagonal of the images)
for the time being, and would be corrected later on through self-
calibration adjustment techniques. The distortion parameters
were disregarded for the time being and initialised to zero. They
would also be precisely computed once the whole image block
had been preliminarily triangulated and an adjustment with self-
calibration could be carried out!
2.2 Datum Definition
Due to the fact that no control point information was given, two
major problems appeared in the orientation procedure. Firstly,
the definition of a global coordinate system (Cartesian XYZ-
system) in which the image block should be defined and the
object be reconstructed in.
For defining this system (seven degrees of freedom) one tie-
point lying close to the motorcycle on the ground was chosen as
origin (X=Y=Z=0). This way the three translation components
of the system were defined. A second point was fixed on the X-
axis of the local system (Y=Z=0). Thus, two rotations were
defined. Finally, a third point was chosen to be fixed, lying in
the horizontal XY-plane (Z=0) in order to define the third
rotation, respectively. It was paid attention that the three points
defining the datum were chosen in such a way, that the Y-axis
was approximately parallel to the motorcycle's symmetric
plane,
The scale - being the seventh degree of freedom - could
unfortunately at this step not really be defined. The only
valuable information regarding the scale, were the distance
between the front and the rear wheel and the brake disks
diameters that were taken from the motorcycle's specifications
sheet. Unfortunately, no homologous points could be found on
the brake disks or at the wheel axes or centres. Hence it was
decided to define a fixed arbitrary distance between two points
provisorily to achieve first adjustment results. A precise scaling
procedure would be carried out^later on using fictitious
Observations!
Since the image configuration was very weak, it was necessary
fo measure a great number of tie points in the overlapping
image areas, but the finding of useable homologous points was
restricted to few limited small areas in the images. Hence
737
additional ‘tie-features’ had to be employed to stabilise the
block and were integrated into a hybrid adjustment (Figure 1).
3. HYBRID ADJUSTMENT AND FICTITIOUS
OBSERVATIONS
When talking of hybrid adjustment, it is meant that the
incorporated input data (observations) can come from widely
different origins (Kraus, 1996). For example, polar points
measured with a tachymeter, spatial directions between pairs of
points measured with an electronic distance-measuring
instrument, GPS measurements, or the well-known image
points. Observations can also be provided by so-called shape or
feature information. Features are (usually) not observed with an
instrument, but are defined by the human senses and
accumulated knowledge. Typical feature information can be:
e Horizontal or vertical planes: one or more points needed
e Straight lines: two or more points needed
e Arbitrary planes: three or more points needed
e Parallel straight lines: three or more points needed
* Parallel planes: four or more points needed
Also, spatial curves can be used instead of straight lines and
spatial surfaces can be used instead of planes. Such spatial
curves and surfaces can allow general relations between (object)
points to. be taken into account. Observations (done in the
mind) leading to feature information are called fictitious
observations (Kraus, 1996).
Every such feature is described in an individual local coordinate
system. In these local systems the observations are
mathematically defined. For example, for image point
observations, that local coordinate system is defined through
the exterior orientation parameters of the image. The observed
values would be the image point coordinates. But also polar
point measurements are defined in local coordinate systems.
Here the unknowns would be the translation and orientation of
the local system, and the observations the azimuth and
horizontal angle, as well as the observed range to a specific
point. Hence all observations types are described in individual
local coordinate systems and are equally treated in the hybrid
adjustment process.
3.1 Lines and Circles as Tie-Features
For nearly all projects in photogrammetry points are used for
tying and controlling the image block. But in some cases no
uniquely identifiable points can be found on the three-
dimensional object, mainly due to lack of texture information.
Nevertheless, often certain identifiable features appear in
multiple images and can be used as tie or control information
(see Figure 1). These features can be lines, curves or circles for
example, or any other analytically describable feature, e.g.
polynomial curves (Kager, 1980) or even splines (Forkert,
1994). If such features cannot be described in a mathematical
way, the problem becomes insolvable.
Figure 4 and Figure 5 should give an impression on the idea of
using non-homologous points to reconstruct features in object
space.
The task is to find the three dimensional objects (L, a line, and
C, a circle)! Considering that the elements of the exterior and
interior orientation are known and no homologous points can be
found on the tie or control element, we measure image
