INVESTIGATION OF GEOMETRIC CONSTRAINTS FOR MATCHING HICH
RESOLUTION SATELLITE IMAGES
P. M. Dare
Airborne Research Australia, Flinders University, PO Box 335, Salisbury South, SA 5106, AUSTRALIA
Paul.Dare@AirborneResearch.com.au
Commission III, WG 11/2
KEY WORDS: Ikonos; DEM/DTM; Matching; High resolution; Satellite imagery
ABSTRACT
The creation of digital surface models from stereo imagery is a well-understood procedure that is central to digital photogrammetric
processing. Recently, however, attention has focused on the creation of surface models from high resolution satellite imagery, which
is not quite so straightforward due to the specific attributes of spaceborne imaging systems, and the fact that some data suppliers do
not release details of the sensor and camera models. This paper describes a matching procedure for creating digital surface models
(DSMs) from stereo imagery acquired by the high resolution Ikonos satellite. Central to this matching procedure are the geometric
constraints that are commonly used to reduce the search space and hence constrain the matching solution. Results are presented of
the use of two different geometric constraints (one image space constraint, and one object space constraint) applied to two very
different Ikonos stereopairs. A range of digital surface models were created, which, when compared to reference data, showed height
differences of less than a few metres. Furthermore, visual evaluation of the resulting surface models showed that both geometric
constraints yielded a good representation of the true surface.
1. INTRODUCTION
The topic of automatic image matching, especially for the
purpose of surface modelling, has received considerable
attention for many years. (For a concise historical summary, the
reader is referred to Samadzadegan, 2002.) Although many
early problems associated with image matching have been
resolved by the development of new algorithms, or the
application of high-end technology, new issues continue to
arise. The majority of previous image matching research, from
a geometric point of view at least, has utilised aerial
photography or moderate resolution satellite imagery (such as
SPOT panchromatic data). Nowadays the wide availability of
high resolution stereo satellite imagery means that image
matching can be thoroughly investigated. The different
attributes of these sensors, as compared to aerial photography,
mean that new problems have to be resolved, and hence new
algorithms or matching strategies have to be developed.
Matching conjugate points in high resolution stereo satellite
imagery is more challenging than in air photos due to the far
more limited opportunities for satellite image acquisition.
Aerial photography that is recorded for the purpose of
topographic mapping would always be acquired under ideal
conditions, namely good illumination, appropriate base to
height ratio for the level of terrain undulation, and correct scale
for the ground features being imaged. With high resolution
satellite imaging these parameters can rarely be changed.
Illumination is dependent upon season and latitude (time of day
is fixed by orbital parameters); base to height ratio is set by the
satellite operator; and, scale is fixed by sensor resolution,
orbital height and look angle.
The two main consequences of the difference between
automatic image matching with high resolution satellite
imaging, and with aerial photography, are firstly that alternative
matching strategies may be required to account for the lack of
sensor orientation information, and secondly that results cannot
be expected to be as good as those from aerial photography (a
result confirmed by Fraser et al., 2001; Fraser et al., 2002a).
This paper presents the results of a study which investigated
different matching strategies for pairs of Ikonos images. In
particular, the study has focussed on methods used to constrain
the search space. As a result, two different geometric
constraints have been evaluated. The first constraint, based on
epipolar geometry, operates in image space, and can be applied
to all images, whether or not they are aligned to epipolar
coordinates. The second constraint is based on the affine
projective model (Fraser et al, 2002b), which like rational
polynomial coefficients (RPCs), maps image space coordinates
to object space coordinates. Matching is constrained by limiting
the search for conjugate points along vertical nadir lines in the
stereomodel. By applying these constraints to two Ikonos
stereopairs, the relative advantages and disadvantages of each
constraint have been compared.
2. THE AFFINE PROJECTIVE MODEL
The affine projective model is somewhat similar to the RPC
model in that it relates image space coordinates to object space
coordinates without any knowledge of the sensor model or
exterior orientation (EO). The general form of the model
describing an affine transformation from 3D object space
(X, Y, Z) to 2D image space (x, y) for a given point i is expressed
as:
xX; = Aj X; + A»Y, + A3L; + Ay
(1)
This model comprises eight parameters per image, these
accounting for translation, rotation, and non-uniform scaling
and skew distortion. Implicit in (1) are two projections, one
International Archi
MC
scaled-orthogonal
resolution satellite
Quickbird with nari
projection is paralle
practical tests to be
the reported implen
model parameters
triangulated ground
photogrammetric bu
3. D/
Two pairs of satelli
stereo pair of Ikonc
geometry by the de
over the city of Mel
does not vary signif
sea level and the h
The Central Busine
area, and contains b
imagery depicts wal
stereopair is shown |
Figure
The second image |
Ikonos stereopair of
covered a wide ;
approximately 7km
region was chosen i
to the Melbourne d
terrain, vegetation |
of this stereopair is <
