The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008
1317
THE AFFINE PROJECTION MODEL AS A TOOL
FOR RAPID GEO-CODING OF IRS-P5 SATELLITE IMAGERY
F. Dadras Javan 3 *, A. Azizi b
a Dept. of Geomatic Engineering, Faculty of Engineering, University of Tehran, Iran - fdadrasjavan@ut.ac.ir
b Centre of Excellence for Disaster Management, Dept, of Geomatic Engineering, College of Engineering, University of
Tehran, Iran - aazizi@ut.ac.ir
KEY WORDS: High Resolution, Pushbroom, IRS-P5, Mapping, Sensor Modeling, Rational Function Model, 3D affine
transformation, Hazards
ABSTRACT:
Nowadays, the information content of the satellite imageries as a means for the disaster forecasting and management has attracted the
worldwide attention more than before. On the other hand, linear array satellite images such as Spot, Ikonos, QuickBird, IRS, etc.,
with their flexibility in acquiring stereo coverage over any part of the globe, have proven to be an excellent replacement for the other
space-borne imaging systems such as digital frame cameras or whiskbroom technologies. The first step for using such data is Geo
coding. High resolution data increase the need for higher accuracy data modeling. Up to now different models with different
accuracy have been discussed. These models are divided into two main groups of the so called rigorous and non-rigorous models.
The rigorous approaches are the most accurate but need crucial data such as satellite ephemeris and inner orientation parameters
which are not always available. The non-rigorous models such as rational polynomials, DLT or 3D affine transformations on the
other hand are less accurate but enjoy the advantage of being independent from the auxiliary information. In line with several other
research works already performed by other researchers, this paper sets its main goal to compare the simple 3D affine model, as a
replacement transformation for the more sophisticated rational function approach. The adopted strategy is based on generating virtual
ground control points using rational polynomials intersection by means of available RPCs. The generated virtual GCPs provide a
reliable data for estimating the degree of fitness of the 3D affine model to the rational polynomial transformation. This paper reports
the result of the tests conducted on a high resolution stereo IRS-P5 satellite image. Other related issues including different methods
for estimating initial values needed for the solution of the rational polynomials intersection, such as DLT, 3D affine and truncated
rational polynomials are also presented and discussed.
1. INTRODUCTION
After the recent Tsunami and earthquake disasters with their
devastating effects, the information content of the linear array
satellite imageries as a means for the disaster forecasting and
management has gained much more importance than before.
One of the crucial preliminary stages after any natural disaster is
the rapid mapping of the damaged areas using satellite
imageries. This process entails a great deal of computations and
field works which hinder the rapid response to the preliminary
mapping demands. Two main approaches are used for geo
coding of linear array imageries. The first approach is the so
called rigorous model. This approach is based on the physical
modeling of the linear array motion and attitude variations.
However, this Method may not be appropriate for rapid
mapping since it requires necessary orbital information as well
as the sensor calibration parameters which may not be
accessible. The second approach uses the rational polynomial
model (RFM) as a replacement for the rigorous method. Again,
the RFM coefficients are included in the metadata and may not
be accessible in all circumstances. The solution of the RFM also
requires the regularization and normalization. The RFM
intersection is solved iteratively and hence demanding initial
values for the object coordinates. Moreover, the solution may
undergo computational collapse for a given dataset. These
complications make these approaches non-optimal for rapid
mapping applications. Taking into account the fact that high
resolution satellite images inevitably have large focal length, it
can be seen immediately that these imageries enjoy a very
narrow field of view. The very small camera field of view
makes the incoming signals almost parallel. This particular
geometry provides a simple linear-parallel relationship between
the image space and the object space and makes a simple eight
parameters affine transformation optimum for geo referencing
applications. Simplicity of the formulation (i.e. only eight affine
parameters for the entire scene), few numbers of required
GCP’s and the achieved accuracy makes this approach very
attractive from the rapid mapping point of view. Nevertheless,
in practice several unpredicted factors my influence the
accuracy of the transformation. One of the major influential
factors in this respect is the terrain relief undulations. This
approach has been already evaluated by different researchers
worldwide and reasonably accurate results have been reported
using only few numbers of GCP’s (Fraser et al., 2004; Fraser et
ah, 2003; Yamakawa et al 2004). The main task undertaken in
this study is to investigate the fitness accuracy of the 3D affine
model with the RFM, as far as the terrain independent scenario
is concerned. The adopted strategy for the evaluation of the
preliminary results is based on the generation of a network of
the so called virtual GCP’s whose coordinates are obtained by
the available RPC’s. Few number of well distributed virtual
GCP’s serve as the reference data to determine the
transformation parameters of the 3D affine transformation and
the rest of the virtual ground points are considered as check
points for the evaluation of the absolute accuracy. All accuracy
figures are presented for the check points in the object space.
This strategy for the accuracy evaluation adopts the accuracy of
the virtual ground points generated by the RPCs as a criterion
for the evaluation of the fitness accuracy of the affine model.
In the sections that follow the basic concepts of the RFM and
the 3D affine models are reviewed first. This is then followed
by the review of the formulations of the RFM intersection. The