he DARPA
, Washing-
d Research
3, Inc.
“ment with
’roceedings
niques with
486, pages
trol for ex-
ing, 36(4),
|, and Least
McKeown,
m multiple
lerstanding
a, Novem-
ency, Mor-
McKeown,
m multiple
mputer Vi-
tle, Wash-
. M. McK-
| matching
| Proceed-
Techniques
ume 2486,
. M. McK-
iade struc-
phics, and
rammetric
In /nterna-
e Sensing,
ation and
s. In Pro-
Workshop,
anced Re-
shers, Inc.
FEATURE-BASED PHOTOGRAMMETRIC AND INVARIANCE TECHNIQUES
FOR OBJECT RECONSTRUCTION
H. F. Barakat, K. Weerawong, and Edward M. Mikhail, Purdue University
Commission III - IWG II/III
KEY WORDS: Invariance, Linear Features, Object Reconstruction, Least Squares Adjustment
ABSTRACT
Linear features, with independent descriptors in the object space, are affectively used in photogrammetric restitution
and object reconstruction.
Point-based invariance is discussed and its applications in Image
Understanding/Computer Vision are contrasted with photogrammetry. Object reconstruction, being a common
invariance and photogrammetric task is evaluated by both techniques using synthetic and real image data. Research
is continuing on multi-image invariance, multi-feature construction, and combined invariance/photogrammetry
techniques.
1. INTRODUCTION
Imagery used to reconstruct the objects recorded is in
general two-dimensional representation of usually three-
dimensional objects. Image features are of three types:
points, lines and areas. Until recently, photogrammetric
methodology has been based primarily on point
features, particularly because of extensive use of hard-
copy image input. The increased use of digital imagery
has opened up opportunities for exploiting linear
features since they are both abundant in imagery of
human infrastructure, and amenable to extraction by
automated algorithms. The inclusion of linear features,
alone or in combination with point features, into
photogrammetric reduction algorithms requires careful
development and analysis.
Image invariance refers to the existence of properties,
derived from images, which are invariant under specific
imaging geometry, the most common of which is central
or perspective projection. One very early property used
in graphical rectification is the anharmonic or cross-
ratio. In recent years, activities in Image Understanding
(IU) and computer Vision (CV) has resulted in
significant development in image invariance. As in
photogrammetric research, point-based development
preceded line-based invariance. Although IU/CV
applications of invariance encompass different tasks,
object reconstruction from overlapping imagery is an
application which is also common to photogrammetry.
Feature-based, particularly linear features,
photogrammetric techniques for the reconstruction of
imaged three-dimensional objects is discussed in section
2. A brief introduction to the invariance concept and its
uses is given in section 3. Results from experiments
using both simulated and real imagery are provided for
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
both techniques. Conclusions and continuing research
are in the fourth and last section.
2. LINE-BASED PHOTOGRAMMETRIC
RESTITUTION
2.1 Linear Feature Description
A point feature is represented by two coordinates in 2D
space and three coordinates in 3D space. Linear
features can be similarly described. Considering
straight lines, they are defined by two parameters in 2D
space, and four independent parameters in 3D space,
expressed by equations (2.1) and (2.2), where p (the
distance from the origin to the line) and a (its angle
with the x-axis) are the 2 parameters in 2D; q (the
distance from the origin to the line), and ,. p, p,
(angles effecting rotation such that the line is along one
coordinate axis) are the 4 parameters in 3D. A circle
is defined by 3 parameters in 2D, and 6 independent
parameters in 3D, and represented by equations (2.3)
and (2.4) in which x, y, are coordinates of its center in
the plane, and 7 its radius; X,Y ,Z_ are coordinates of
the center; R its radius, and a, a, are the angles
defining the unit vector P perpendicular to the plane of
the circle in 3D.
X,cos 0. +y sina =p (2.1)
535