a
a SN
RRS
FR
————M
TRANSFORMATION
ULTRA-PRECISE MEASUREMENT USING AFFINE
ATSUSHI OKAMOTO
KYOTO UNIVERSITY KYOTO JAPAN
COMMISSION V
ABSTRACT
Affine transformation plays a minor role in pho-
togrammetry. However, this transformation shows
great effectiveness in the cases where projective
transformation is very weak. This paper presents a
general orientation theory of two-dimensional
affine images and clarifies characteristics of the
model construction and the one-to-one correspon-
dence between the model and object spaces. Next,
a general free network theory for affine images is
constructed with twelve linearly independent vec-
tors. Practical characteristics of the proposed the-
ories are discussed through tests with simulated ex-
amples.
INTRODUCTION
Affine imageries do not exist in reality. For this
reason the orientation theory of affine images has
probably not been derived until now. However,
this orientation theory is of great practical use in
many photogrammetric fields such as precise
three-dimensional measurement of small objects
using a conventional comparator and the analysis
of satellite CCD camera imageries. In the latter
case the conventional orientation approach of pho-
tographs cannot be employed, because the field an-
gle of the CCD camera is very narrow and height
differences in the photographed terrain are very
small for the flying height of the satellite. There-
fore, this paper derives a general orientation the-
ory of two-dimensional affine images and clarifies
the geometrical properties of the relative and abso-
lute orientation. Then, a general free network the-
ory for affine images is constructed. The theories
proposed here are tested with simulated examples
so as to explore the difficulties when applying them
to practical cases.
GENERAL ORIENTATION THEORY OF TWO-
DIMENSIONAL AFFINE IMAGES
Let a three-dimensional object space (X,Y,Z) be
projected into a plane based on affine transforma-
tion(See Figure-1.). The basic equations relating
measured image plan
(affine image )
object space
Figure-1 : parallel projection of an object space
into the measured plane of the compa-
rator coordinate system.
an object point P and its measured image point
Pe(Xc.ÿc) are described (Okamoto (1989)) as
Xe = a1X + a2Ÿ + a3Z + a4
(1)
yc = asX + agY + a7Z * ag
in which ai (i=1, --,8) are independent coefficients.
Geometrically, the eight orientation parameters of
the affine image are considered to be three rotation
parameters (9,9K) of the image, two translation
elements (Xoc, Yoc) which indicate two of the three-
dimensional coordinates of the origin of the mea-
sured image coordinate system (Xc;ÿc) with respect
to the object space coordinate system (X,Y,Z), the
image scale s, and two rotation parameters (&,B)
describing the relationship between projected rays
and the normal to the image plane. The eight ori-
entation parameters of a single affine image can
thus be provided uniquely if four control points
are available.
Next, we will consider the orientation problem of a
stereopair of two-dimensional affine images (See
Figure-2.). The basic equations are written down
as