ORTHORECTIFICATION OF HIGH RESOLUTION SATELLITE IMAGES 
P. Boccardo?, E. Borgogno Mondino**, F. Giulio Tonolo*, A. Lingua" 
? Politecnico di Torino, Dipartimento di Georisorse e Territorio, C.so Duca degli Abruzzi 24, 10129 Torino, ITALY 
(piero.boccardo, enrico.borgogno, fabio.giuliotonolo, andrea.lingua)@polito.it 
KEY WORDS: Remote Sensing, Orthorectification, Rigorous, Orientation, High resolution, , Accuracy 
ABSTRACT: 
The growing availability of high resolution satellite images leads to evaluations that are aimed at the definition of the mapping scc 
they can reasonably be defined for. The problems connected to the use of satellite images for the production of orthophotos in urban 
areas at a middle scale are dealt with in this work. An analysis of the residuals perspective errors connected to the presence of great 
altimetric discontinuites due to buildings and infrastructures is made. The effects of the use of a Dense DTM are evaluated and the 
problem of displacements on orthophotos induced by raised volumes are considered too. 
A procedure has been developed which allows the orthoprojection of high resolution satellite images using a Dense DTM and non- 
parametric models. In particular non parametric algorithms based on Rational Function Models (RFM) and Neural Network (Multy 
Layer Perceptron) have been implemented. À multi-image (along-across track stereo images, multi-temporal images) and/or multi- 
sensor approach can be followed for the production of true orthophotos: this approach can prevent from the problem of data 
duplication next to elevation discontinuites. 
The most frequently used non-parametric methods are based on 
3D rational polynomials, and which in literature are known as 
The introduction of high and very high resolution satellite the Rational Function Model, RFM — Rational Polynomial 
images has made it necessary to revise the geometric correction Coefficients. Rational Polynomial Camera, RPC — Rational 
techniques that are used in this field. There has been a Function Coefficients, RFC (Dowman, Tao, 2002). Moreover a 
transition, on the basis of evaluations aimed at the definition of new prototype of a geometric correction procedure based on a 
map scale for which they can reasonably be defined, from Multy Layer Perceptron type (MLP) neural network, has been 
simple 2D polynomial models to rigorous or non rigorous 3D proposed in this paper. These latter methods have been analysed 
models derived from digital aerial photogrammetry. and verified in the application field. 
When the investigated zone is an urban area or when the 
territory is characterised by discontinuities, a classical type of 2.1 Rational Function Model 
orthoprojection could be insufficient for mapping purposes and : 
might need to be substituted with a more rigorous approach The rational function is the most commonly used non- 
parametric model, which is implemented in almost all software 
packages for the processing of satellite images. This type of 
approach is used by image salesmen to allow the final user to 
obtain added value products, such as orthoprojection without 
1. INTRODUCTION 
(true orthophoto). 
2. GEOMETRIC CORRECTION 
The geometric correction of high-resolution satellite images can the necessity of having a model of the sensor, but by only 
be carried out using two different approaches: rigorous attaching the coefficients of the relation between the image 
modelling or non-parametric modelling. coordinates and the ground coordinates. 
Rigorous models are based on collinear equations (Toutin, The rational function model allows a relationship to be 
2004) that are adapted to pushbroom acquisition technique determined between the image coordinates (£, ) and the 3D 
which is used by all high resolution satellites. 
In this case, the orientation parameters are modelled as time 
dependent polynomials of a higher degree than the first: the 
coordinates of the object (X, Y and Z) through polynomial 
relations, as shown in (1) : 
estimation of the unknowns requires approximated initial values 
which are extracted from the metadata files usually supplied é= P,(X,Y,Z) : 
together with the images. PX. 1 27) (D 
However the Companies that distribute images are not always 
willing to supply detailed technical information to the final ) = PO Y 7) 
users concerning the platform that is used or about the PALY.Z) 
characteristics of the sensor that are necessary to implement 
rigorous models. It was for this reason that non-parametric where P,, P,, P., P are usually maximum degree polynomials 
models, or rather generalised models (independent of both the equal to 3, corresponding to 20 coe ficients: which can be 
> of sensor and of the acquisition method) were introduced. : 
typeef sensor a e aed ) expressed through equation (2)or 3) 
  
* Corresponding author. 
30 
internatior 
BO Áy.. 
+a, XY + 
Equations 
allow the 1 
coordinate 
estimation 
0+19), it i 
afier havir 
implement 
as to be 
commercié 
Some crit 
First of al 
(truncatior 
normalise 
the interva 
denominat 
in functior 
range, it | 
square sys 
the norma 
polynomi: 
often does 
If we pre: 
users, it is 
conventio; 
homologo 
GPS surve 
distributio 
a numeri 
process co 
Tikhonov 
regularisal 
systems: i 
^? to the € 
conditioni 
square sys 
The choic 
fact obtai 
while van 
the rejects 
The nume 
even thot 
orthoproje 
are conne 
probably « 
verify the 
having to 
has to be 
polynomi: 
implemen