Fraser, Clive
HIGH-RESOLUTION SATELLITE IMAGERY: A REVIEW OF METRIC ASPECTS
Clive S. Fraser
Department of Geomatics, University of Melbourne
Melbourne, Vic 3010, Australia
¢.fraser@eng.unimelb.edu.au
Invited Paper, Working Group VII/3
KEY WORDS: high-resolution satellites, sensor orientation, satellite triangulation, satellite imagery, accuracy aspects,
rational functions, affine projection
ABSTRACT
With the era of commercial high-resolution earth observation satellites having dawned with the launch in September, 1999
of Tkonos-2, it is imperative that there is an understanding in the photogrammetric and remote sensing communities of the
full metric potential of high-resolution satellite line scanner imagery. This paper discusses the metric exploitation of 1m
satellite imagery, and specifically looks at the options available for multi-image restitution in situations where optimal
ground point triangulation accuracy is sought. Such applications would include automated DTM generation and high-
accuracy feature determination for both mapping and geopositioning. The paper commences with a brief discussion of
accuracy aspects and then overviews the collinearity model for orientation/triangulation of line scanner imagery. Alternative
restitution models are then reviewed. These include rational functions, the direct linear transformation and an affine
projection approach. Each restitution model has its merits and limitations, and of central importance is the provision or lack
of provision of the ‘camera models’ for the different high-resolution satellite systems. With the prevailing level of
uncertainty over just what critical sensor calibration information will be made available by the satellite imaging companies,
there is consequently a need for a range of alternative, practical approaches for extracting accurate 3D terrain information
from 1m satellite imagery.
1 INTRODUCTION
In September 1999, with the successful launch of the 7konos-2 satellite by Space Imaging, the photogrammetry and remote
sensing communities entered the era of commercial high-resolution earth observation satellites. One of the great promises of
Ikonos, and its planned competitors such as Quickbird and Orbview III (e.g. Fritz, 1995) is that 1m satellite imagery will
display the metric quality to support topographic mapping to large scales, and even to scales of larger than 1:10,000, as well
as ground feature determination from multispectral imagery to better than 5m accuracy. In order to meet the necessary metric
accuracy specifications, appropriate mathematical models and computational procedures will be required. Although there
has been more than a decade of experience gained with exterior orientation (EO) determination and subsequent ground point
triangulation of line scanner imagery, metric exploitation of 1m satellite imagery will bring with it some new challenges.
These are unlikely to relate to accuracy requirements alone. Instead, much will depend upon the provision or lack of
provision of the necessary sensor calibration model and precise satellite ephemeris data to support optimal multi-image
restitution.
In the past 15 years or so there has been a good deal of research attention paid to the recovery of 3D cartographic
information from satellite line scanner imagery. Initial impetus was provided by the formulation of mathematical models to
support both batch and on-line triangulation of cross-track SPOT imagery (e.g. Westin, 1990; Kratky, 1989). This was
followed by developments in 3-line image restitution for systems such as the German MOMS-02 satellite sensor (Ebner et
al., 1992) and refinement of sensor orientation models for the Indian IRS-1C/D satellite line scanners (Radhadevi et al.
1998). In broadly summarising the result of these endeavours, it could be said that under ideal conditions of high-quality
image mensuration and ground control/checkpoints, coupled with favourable imaging geometry (e.g. base-to-height ratio of
0.8 or more) and provision of sensor calibration data, ground point determination to 0.3 pixels is possible (Ebner et al.,
1996), whereas accuracies of between 0.5 and 2 pixels are more commonly encountered in practical tests. Thus, for SPOT
452 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000.
