(1987), Chen, Huang (1990) and Zhang, Faugeras
(1991). The use of surface data (digital terrain
models) as control information in block triangulation
and absolute orientation was treated by Ebner, Strunz
(1988) and Rosenholm, Torlegärd (1988).
In this paper a systematic overview concerning the use
of points, lines and surfaces for photogrammetric
orientation and object reconstruction tasks is given. It
is shown, which tasks can be performed, and the
respective minimum number of features needed for a
solution is derived. Some remarks on degenerate
configurations are given and the evaluation of
precision using general geometric features is shown by
means of simulated examples.
2. MINIMUM CONFIGURATIONS
In the following first some basic considerations
concerning the use of points, lines and surfaces for
photogrammetric orientation and object reconstruc-
tion tasks are given, and then the minimum number
of general geometric features needed for the solution
is derived. It is assumed that the extraction of features
and the precise determination of their locations in
image space has already been accomplished and that
the correspondences between these features are estab-
lished, which in general are non-trivial problems.
Furthermore, it is supposed that for the orientation
tasks approximate values of the unknown parameters
are available, so that linear models can be derived
from the nonlinear models and the problem is solved
iteratively. Linear algorithms as such are not
discussed.
The general method for image orientation and object
reconstruction is based on a bundle block adjustment,
where the orientation parameters of multiple images
and the coordinates of the object points are estimated
simultaneously. Basic orientation tasks, which are
implicitly comprised in multiple image orientation,
can be classified into space resection (2D-3D), relative
orientation (2D-2D), and absolute orientation (3D-
3D). The mathematical model used in 2D-3D and 2D-
2D orientation is based on perspective projection,
which can be formulated by the collinearity equations,
the underlying model of 3D-3D orientation is a spatial
similarity transformation.
Lines in 2D image space can be described in several
forms, e.g. parametric or implicit. If we assume a line
to be represented by the image coordinates of a
specific number of points, which uniquely define the
line, then the collinearity equations can be used for
points as well as for lines. In this paper only two types
of algebraic lines are considered: straight lines and
circular lines in object space, which in general
correspond to straight lines or ellipses in image space.
Basically the image lines can result from the
114
perspective projection of 3D lines or from occluding
contours of 3D surfaces, the 3D positions of which
depend on the exterior orientation of the images.
However, these contours can not be used for 2D-2D
correspondence. According to the specific orientation
task the lines and surfaces can serve as "tie" features
or as control features. This means that the spatial
position and orientation of a particular feature is
either unknown and has to be determined in the
course of the orientation process or is given as control
information in object space.
In the following for the basic orientation tasks the
minimum number of corresponding features is
derived, which is necessary to yield a solution.
Degenerate configurations are not considered. An
overview is given in Tab. 1 - 3, where n is the number
of observation or condition equations and u is the
number of unknowns.
2.1 Space resection (2D-3D)
Space resection involves the determination of the
exterior orientation of a single image, and is described
by 6 parameters.
Points in image space:
A point is defined by 2 coordinates in image space,
which are given as observations, and 3 coordinates
in object space, which are unknowns.
- Point to point correspondence:
If the corresponding feature in object space is a
given point, its 3 object coordinates are known.
Therefore, 3 image points and their corresponding
object points are necessary to determine the
exterior orientation parameters. The solution is
based on fitting the bundle of rays to the corre-
sponding object points.
- Point to line correspondence:
If the corresponding feature in object space is a
given line, 2 conditions are imposed on the 3
unknown object coordinates. Consequently, 6
points in image space and their corresponding
lines in object space are needed. The solution is
based on fitting the bundle of rays to the corre-
sponding line features in object space.
Lines in image space:
A straight line is uniquely defined by 2 parameters
(observations) in image space and 4 parameters
(unknowns) in object space. An ellipse is defined
by 5 parameters (observations) in image space and
in case the corresponding feature is a circle in
object space it is described by 6 (unknown)
parameters.
- Line to line correspondence:
Straight line: If the straight line is given in object
space, its 4 line parameters are known. Therefore,
3 straight lines in image space and the correspon-
re
th
ur
de
In
is