INDIRECT EPIPOLAR RESAMPLING OF SCENES USING PARALLEL PROJECTION
MODELING OF LINEAR ARRAY SCANNERS
M. Morgan“, K. Kim”, S. Jeong”, A. Habib”
* Department of Geomatics Engineering, University of Calgary, Calgary, 2500 University Drive NW, Calgary, AB,
T2N 1N4, Canada - (mfmorgan@ucalgary.ca, habib@geomatics.ucalgary.ca)
b Electronics and Telecommunications Research Institute (ETRI), 161 Gajeong-Dong, Yuseong-Gu, Daejeon, 305-350,
Korea — (kokim, soo) @etri.re.kr
PS WG I11/1: Sensor Pose Estimation
KEY WORDS: Photogrammetry, Analysis, Modelling, Method, Pushbroom, Sensor, Stereoscopic, Value-added
ABSTRACT:
Linear array scanners are used as a substitute of two-dimensional/frame digital cameras since they can produce high-resolution
digital images comparable to scanned aerial photographs. On the other hand, digital frame cameras have inadequate size that is
dictated by technical considerations/limitations. In general, rigorous sensor modelling involves the description of the physical
process of data capture using such a sensor. For imaging systems, rigorous modelling incorporates the system's interior and exterior
orientation parameters. Such parameters might not be always available for linear array scanners (e.g., commercially available
IKONOS scenes). Deriving these parameters requires numerous ground control points. Moreover, the estimation process is
geometrically ill posed due to the narrow angular field of view of the imaging system. Recently, parallel projection has emerged as
an approximate model (for high altitude scanners with narrow angular field of view) that can be used to represent the mathematical
relationship between scene and object space coordinates using few parameters. This paper outlines the derivation of resampling
approach of linear array scanner scenes according to epipolar geometry. Conjugate points should have no y-parallax in the resampled
scenes. Moreover, they should have an x-parallax that is linearly proportional to the corresponding object height. Such requirements
can only be met by projecting the original scenes into a common horizontal plane. The paper explains the selection of such plane to
meet these specifications. Experimental results using IKONOS data demonstrate the feasibility of the approach.
1. INTRODUCTION
Images generated according to epipolar geometry have the
prime characteristic of having no y-parallax values. This fact
makes them important prerequisite for many photogrammetric
applications such as: automatic matching, automatic relative
orientation, automatic aerial triangulation, automatic DEM
generation, ortho-photo generation, and stereo viewing.
Two-dimensional digital cameras with large number of pixels
comparable to scanned aerial photographs are not yet available.
Therefore, linear array scanners are introduced to overcome this
limitation by obtaining successive narrow coverage on the
ground. Linear array scanners can be space-borne such as
IKONOS, SPOT and MOMS, or airborne such as ASD-40. The
high resolution and large coverage obtained by these scanners
motivate their exploitation in performing photogrammetric
activities, which could bring challenge to traditional
topographic mapping with aerial images (Fritz, 1995).
In order to achieve highly accurate products, accurate geometric
modelling has to be adopted. Two main categories of sensor
modelling exist; rigorous and approximate modelling. The
former describes the true geometry of the image formation and
thus it is the most accurate model. Therefore, it has been
frequently used in many applications (Lee and Habib, 2002;
Habib et al, 2001; Lee et al, 2000; Wang, 1999; Habib and
Beshah, 1998; McGlone and Mikhail, 1981; Ethridge, 1977).
However, many ground control points are required for
estimating the parameters associated with the rigorous model. In
addition, the indirect determination of these parameters for
space-borne scenes can be instable, due to the narrow angular
field of view (Wang, 1999). Furthermore, the parameters
associated with the rigorous model can be concealed by the
scene provide (e.g., IKONOS), As a result, other approximate
models exist including the rational function model, the direct
linear transformation, self-calibrating ^ direct ^ linear
transformation and parallel projection (Tao and Hu, 2001; Ono
et al., 1999; Wang, 1999; Abdel-Aziz and Karara, 1971).
Among these approximate models, the parallel projection model
is adopted and utilized for epipolar resampling of linear array
scanner scenes. Section 2 briefly discusses the epipolar
resampling of frame cameras and introduces the linear array
scanners. The reasons for choosing the parallel projection model
and its mathematical formula are discussed in Section 3. Section
4 is dedicated for developing the epipolar resampling approach
of linear array scanner scenes. Experimental results using
IKONOS data are presented in Section 5. Finally, Section 6
includes the conclusions and recommendations for future work.
2. BACKGROUND
2.1 Epipolar Resampling of Frame Images
The main objective of epipolar resampling is to generate
normalized images, which have the innate property that
corresponding points lie on the same rows. Epipolar resampling
of frame images is a straightforward process (Cho et al., 1992).
Figure 1 depicts two frame images at the time of exposure
together with the normalized images. The resampling process
necessitates projecting the original images onto a common
International Ar
plane, whose
parameters of
normalized scer
case, an object
normalized ima
normalization [
frame images, it
linear array sca
first, some term
their geometric 1
Left original ima
0
[Left normal
Figure
2.2 Difference:
It is important |
and “scene” thr
Figure 2.
Figure 2. À sequ
An image is de
with one expos
contains only or
complete image.
many 1-D imag
station. The ma
object space anc
the collinearity
Parameters, EOI
appears).
In contrast, a sce
one (as in frame
array scanners)
short single trip
image and scen
Scanners, the sce