1293
EPIPOLAR LINE GENERATION FROM IKONOS IMAGERY
BASED ON RATIONAL FUNCTION MODEL
Dan Zhao 3 , Xiuxiao Yuan 3 , Xin Liu 3
a School of Remote Sensing and Information Engineering, Wuhan University,
129 Luoyu Road, Wuhan 430079, China - zhaodanpppp@163.com
Commission IV, WG IV/9
KEY WORDS: Remote Sensing, IKONOS, Space Photogrammetry, Accuracy, Geography
ABSTRACT:
High-resolution satellite imagery is expected to be a major source of 3D measurement of ground, especially automatic stereo
plotting to generate DEM by stereo matching technology is highly required. Epipolarity is a very useful concept in processing stereo
imagery, and it has been widely used in process images captured by frame cameras. Different from perspective image, every scan
line of linear array scanner scene has different perspective centre and attitude; this has limited the use of epipolar theory in
processing the linear array scanner scenes. The purpose of this paper is to develop a method which can be used to generate the
approximate epipolar line of this kind of imagery. The paper firstly describes the epipolar geometry of linear array scanner scenes;
then explains the rational function model (RFM) and proposes the use of RFM to generate the epipolar lines; finally, makes a
accuracy assessment of the rational polynomial coefficients (RPCs) used in experiment and emphatically describes the procedure
and experiment of the generation of epipolar lines and epipolar line pairs of IKONOS stereo imagery.
1. INTRODUCTION
1.1 General Instructions
High-resolution satellite imagery captured from linear array
scanner is valued for their great potential in stereo plotting. The
linear scanner with up to one-meter resolution from commercial
satellites could deliver more benefits and provide a challenge to
traditional topographic mapping based on aerial image. It is
expected to be a major source of 3D measurement of ground.
Due to the complicated imaging geometry of high-resolution
satellite imagery, many traditional theories of photogrammetry
can not be applied to this kind of imagery directly, so the
appearing of high resolution satellite imagery provide new
research contents to the space photography.
Epipolarity is an important concept in processing stereo images.
A useful property of an epipolar line is that all corresponding
image points lie on the corresponding epipolar line pairs. This
property makes epipolar constraint as an important constraint
condition for image matching. Many existing stereo matching
algorithms use this constraint to confine search dimensions,
reduce processing time and achieve reliable match estimates
(Zhang et al., 1995; Kim, 2000). For the aerial and perspective
image, epipolar geometry has been well founded and widely
used. But due to the complex imaging geometry of linear array
scanner scene, this geometry of such imagery is not easy to
obtain and it has not been used as widely as it used in
processing perspective images. To resolve this problem, many
scholars have made a lot of efforts. Some have assumed that
epipolar geometry would be the same for push-broom imagery
as perspective image, some do not use this geometry at all. This
paper is aims at developing a method which can be used to
generate the epipolar lines of the linear array scanner scenes.
It is important to select an appropriate sensor model to be
utilized before the study of the epipolar geometry. The rigorous
and generalized sensor models are the two broad categories of
sensor model in use (McGlone, 1996). The representative of
rigorous model is the collinearity equations. The generalized
models include RFM, two-dimensional affine model, direct
linear transformation (DLT). The sensor model used in this
paper is RFM. It is a generalized model which can be generated
from the physical sensor model and substituted for all sensor
models and it is capable of achieving high approximation
accuracy. However, RFM has a complex mathematical
expression. Therefore, it is difficult to generate the exact
epipolar curve for linear array scanner scenes use RFM. In this
paper epipolar lines are generated by approximate method.
1.2 The Epipolar Geometry of Linear Push-Broom Scenes
For perspective image, there is only one perspective centre for
the entire image, so the epipolar curve can be defined as the
intersection between the epipolar plane and the image plane.
The epipolar curve of any perspective images can be always
represented as straight line, which is a well-known property of
the epipolar geometry of perspective image. Conjugate epipolar
pairs are exist for perspective image, the conjugate points must
lie on the conjugate epipolar lines. Different from perspective
image, every scan line of linear array scanner scene has
different perspective centre and attitude. It is impossible to
define an epipolar plane as perspective image, so it is not easy
to give a rigorous definition for epipolarity of linear array
scanner scenes.
There are two difficulties in deriving the epipolar geometry of
linear push-broom scenes, the first difficulty is that a generally
accepted mathematical method to describe the relationship
between image and object space has not yet been established
(McGlone, 1996), and the second difficulty comes from the
complexity of such a geometric relationship and difficulty in
expressing the relationship in mathematical forms.