MIDDLE SCALE MAPPING OF CULTURAL SITES 
BASED ON HIGH RESOLUTION SATELLITE IMAGES 
P. Boccardo J ‘ , E. Borgogno Mondino a , F. Giulio Tonolo a 
a Politecnico di Torino - Dipartimento di Georisorse e Territorio 
C.so Duca degli Abruzzi, 24- 10129 Torino - ITALY 
Piero.boccardo@polito.it; enrico.borgogno@polito.it; fabio.giuliotonolo@polito.it 
KEY WORDS: Satellite remote sensing, Orthorectification, Mapping, Land cover, Rational Function Model, Pan-sharpening, 
Classification, High resolution 
ABSTRACT: 
The availability and variety of high resolution satellite images (Eros, QuickBird, Ikonos, Spot5 supermode) have led us to consider 
the possibility of updating middle scale cartography. This paper presents a methodology to produce accurate orthocorrected images 
which can be used to generate updated maps at a scale of 1:10000. Even though commercial software already allow such operations 
to be made, we have implemented an alghorithm that is based on the RFM (Rational Function Model) in order to control the 
positioning errors and the influence of the ground control point distribution on the results. Such an approach also permits us to 
investigate whether any of the 78 RPCs (Rational Polynomial Coefficients) are negligible. We assume that a DEM (Digital Elevation 
Model) of the investigated area is available (e.g. from SAR interferometry, SPOT5 mission,...). In order to define the maximum 
obtainable map scale (which depends on the geometric resolution of satellite images and on the adopted sensor model) accuracy tests 
have been carried out using a reasonable number of check points: the results are presented in this paper. The spectral information 
from multispectral data (XS, with a lower geometric resolution than the panchromatic band, PAN) can be used to produce thematic 
maps which can be considered as an added value of the process, especially in sites of cultural value. A Pan-Sharpening algorithm has 
been implemented paying particular attention to the file size management (through a partitional approach) and the corrispondences 
between PAN and XS spectral ranges. An example of a land cover map that was produced through a neural classification of such 
orthocorrected images is presented. 
1. IMAGE ORTHOPROJECTION 
Cultural site investigations require updated middle scale 
cartography to refer to. High resolution satellite images could 
be succesfully used to face such needs, but orthoprojection has 
to be taken into consideration. In literature remotely sensed 
satellite images have been orthocorrected using both rigorous 
sensor models and general non parametric models. 
1.1 The Rigorous Sensor Model 
Rigorous sensor models are based on collinearity equations; 
each line of the image, but not the entire image, can be 
considered a central perspective. Collinearity equations have to 
be modified to take into consideration the time dependence of 
the sensor’s attitude and position. Starting values can be 
obtained from ancillary data, provided with images, by the 
reseller company: the position vectors, attitude values and the 
starting/end times of acquisition are usually provided. 
Otherwise, the starting values have to be calculated using linear 
models (i.e. time dependent linear transformed DLT equation). 
A rigorous sensor model is often not available; this means that 
general projecting (non parametric) algorithms have to be used. 
1.2 The Rational Function Model 
One of the most commonly used non parametric models is the 
Rational Function Model (RFM). This model, proposed by 
OPENGIS consortium relates the image coordinates (>,0) and 
the three dimensional terrain coordinates {X, Y,Z) through a ratio 
of the 3 rd order (maximum) polynomials (20 coefficients), as 
shown: 
p„(xj,z) 
P h (X,Y,Z) 
P C (X,Y,Z) 
PAXJ,Z) 
(la) 
These equations are called RFM upward equations. It can be 
useful to calculate the terrain coordinates from the knowledge 
of the image coordinates and of the Z values, according to the 
following equations, which are known as RFM downward 
equations: 
x P&,ri,Z) 
P b (4,V,Z) 
y P&.TI.Z) 
Pté.n.Z) 
(l.b) 
In order to correctly estimate polynomial coefficients it is 
necessary to collect a sufficient number of Ground Control 
Points, ranging from a minimum of 7 (1 st order polynomial) to a 
maximum of 39 (3 rd order polynomial), considering that the first 
coeffincient in the denominators is assumed to be equal to 1. 
Both the rigorous and the non parametric models are available 
in commercial software. A non parametric approach, which is 
based on a self developed RFM alghoritm (IDL programming 
language), is shown in this paper. The self developed 
alghorithm permits us to keep the behaviour of such a 
methodology under control.