mmm
K
).01
).02
).05
).08
).03
).00
).00
.00
02
01
01
.00
.00
.00
).03
027
Sr SS el
anbul 2004
ADVANCED INFORMATION EXTRACTION FROM NON-METRIC IMAGES USING
ADAPTABLE ALGORITHMS EMBEDDED IN A HYBRID ADJUSTMENT
G. Vozikis, H. Kager, P. Waldhaeusl
IPF, Vienna University of Technology, Gusshausstrasse 27-29, 1040 Vienna, Austria
[gvozikis, hk, pw]@ipf.tuwien.ac.at
KEY WORDS: Photogrammetry, Adjustment, Calibration, Parameters, Accuracy, Close Range, Feature, Reliability
ABSTRACT:
During the past decades many computer programs have been developed for solving standardized problems in the field of close-range
photogrammetry. Most of these software packages are designed for common tasks and hence algorithmically not very flexible. When
facing the problem of dealing with very special, individual cases in photogrammetry, programs with very versatile mathematical
background and the ability to adopt to very specific situations are needed, e.g. for evaluating a complex traffic accident scene as
described in this paper. At a street intersection a car was hit sidewards by a motorcycle. The task was to calculate the deformation
angle between the upper and the lower section of the fork of the front wheel as precise as possible where a sharp bend occurred
through the crash. This information would be valuable for the technical expert in order to compute the motorcycle's collision speed.
Considering that only four of the available non-metric images were useful and that no calibration, control, or interior orientation
parameters were available, this task became rather complex and scientific. Unfortunately, the image-configuration was very weak
(bad intersection quality of the rays of tie points) and finding of useable homologous points was restricted to few limited small areas
in the images. That is also why geometric tying features had to be introduced. Furthermore, to solve and stabilize the block one had
to employ advanced tools of photogrammetry: shapes (features) in form of planes as well as second order surfaces were introduced.
In addition, some of the unknowns had to be fictitiously observed.
ZUSAMMENFASSUNG:
Wihrend der letzten Jahre wurden viele Computerprogramme entwickelt, um standardisierte Probleme im Gebiet der terrestrischen
Photogrammetrie zu lósen. Die meisten dieser Software-Pakete wurden für Alltagsapplikationen erzeugt und sind deswegen
algorithmisch nicht besonders flexibel. Wenn man sich mit sehr speziellen, individuellen Problemen auseinandersetzen muss, braucht
man Programme mit einem sehr wendigen und vielseitigen mathematischen Hintergrund, die die Möglichkeit besitzen sich an
bestimmte Situationen anzupassen, z.B. um ungewóhnliche Fragestellungen bei einem Verkehrsunfall zu beantworten (wie in dieser
Arbeit beschrieben). Bei einer Kreuzung wurde ein Auto seitlich von einem Motorrad gerammt. Die Aufgabe bestand darin, den
Deformationswinkel eines Knicks der vorderen Motorradgabel zu berechnen, der wáhrend dem Unfall aufgetreten war. Aus dieser
Information konnte ein Gutachter anschlieBend die Kollisionsgeschwindigkeit berechnen.
Es waren vier Amateuraufnahmen vorhanden, aber keine Kalibrierungs- oder Einpassinformation oder Informationen bezüglich der
inneren Orientierung, daher wurde diese Aufgabe relativ komplex und wissenschaftlich. Leider war auch die Bildkonfiguration
ziemlich schwach (schlechte Schnittqualitát der Strahlen) und das Auffinden homologer Verknüpfungspunkte reduzierte sich auf
wenige, kleine Regionen in den Bildern. Deshalb mussten geometrische Verknüpfungs- und Passgestalten eingeführt werden. Um
den Block zu lösen und zu stabilisieren, wurden auferdem erweiterte photogrammetrische Methoden angewandt: Gestalten in Form
von Ebenen und Flächen zweiten Grades. Zusätzlich wurden manche der Unbekannten fiktiv beobachtet.
1. INTRODUCTION (metric camera) or the corners of the images (non-metric
camera) in order to calculate a common interior orientation
When facing the problem of evaluating a traffic accident, close- describing the relationship between camera-space and image-
range photogrammetry was and is the most commonly used space.
solution, since it is a well established, proven and accurate
method. Most of the times the demands are limited and consist
of the determination (location, extents and orientation) of skid
marks, the exact location of wreckage, or in computing the final
positions of the involved vehicles. In general, a number of
control points (that are visible in the imagery) are accurately
measured by means of various surveying techniques and are
needed for the orientation process of the images. Some times
these images are taken with a metric camera, hence the interior
orientation parameters (and distortion coefficients) are known,
which simplifies further processing steps.
The normal procedure begins with the scanning of the
developed film of the images (if not acquired with a digital
camera), measuring the coordinates of the fiducial marks
Next task would be to measure tie points. These homologous
points must be visible in at least two images and are important
for 'tying' the block. Control points can also be used as tie
points. Before performing the first adjustment it is important to
define approximate values for the exterior orientation
parameters, this can either be done by a manual process, or by
using control point information. E.g. if four (or more) control
points are visible in one image the method of Mueller-Killian
(Kraus 1996) could be applied to get the approximate position
and orientation of the new image under investigation. When the
results of the adjustment are accepted, we consider the image-
block as oriented (relation between object and image space
established), which means that we easily can digitise a point in
two (or more) images in order to get its coordinates in the
