)NES
e-mail:
k, RFM
c features
'esence of
ssification
isk to rest
k for both
tric) tests
n is based
1 problem
the strong
ir average
tonomous
‘on neural
n order to
d with an
the flying
1e MIVIS
ds belongs
| infrared,
ctrometers
he Earth’s
he visible
red (1.15-
10 in the
ranges can
1 sciences,
1e territory
icerned, if
ition have
analyzing
uantitative
f. unstable
Band (nm)
20
50
9
340-540
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004
1.2 Available images and software elaboration
The geometric correction operations were performed on two
MIVIS images from an area of the Susa Valley, Piemonte
Region in Italy. They were taken by flights, transversal to the
valley, as shown in figure 1.
p
Figure 1 -Test area for the MIVIS images acquired.
The test area is characterized by a notable elevation variability:
from quota 1000 m a.s.l. along the valley to a maximum quota
above 3000 m a.s.l., sometimes with steep areas (see figure 2).
Such situation can widely represent a generic example of a
mountainous area.
Official orthoimages and the Piemonte Regional Map (scale
1:10000) of the area have been used as cartographic references
for the collimation of the GCPs and of the Check Points (CPs)
and for operating a qualitative analysis of the geometric
correction results.
At last, the DEM of the study area has been used, grid 50m x
Figure 2 — 3D view of the Digital Elevation Model of MIVIS
test zone: the figure shows the strong orographic variability.
50m, for the extraction of the elevation information of the GCPs
and CPs and as auxiliary data required by the RFM and NN
correction methods.
RFM approach has been carried out by the commercial software
OrthoEngine PCI Geomatica 8.2 software. The experimental
method based on NN has been instead implemented in IDL
(Interactive Data Language) language as far as data preparation
and orthoimage generation is concerned, and in MATLAB
language for neural network training and network adoption for
the estimation of image coordinates from terrain coordinates as
successively shown.
873
M I VIS IM M A G ES
MIVIS 1 MIVIS 2
Acquisition data 26/07/1999 26/07/1999
Acquisition time 11:45 11:52
Stich Hei 17100 ft (5212 m 17500 ft (5334 m
Flight Height #15 AST)
Rows number 4000 4001
Columns number 755 755
Spectral Resolution 102 bands 102 bands
Geometric Resolution 4-8m 4—8m
Preprocessing level 0 0
Table 2 — Test MIVIS images features.
1.3 Orthoprojection problems
When dealing with territorial applications it is always important
to correctly approach the scale mapping problem. This means
that ground objects positioning must be coherent for all the used
data (often coming from different sources and reference map
systems). Such problem can be easily solved with geocoded
data such as ancillary and cartographic ones. Not so easy is to
face the problem of MIVIS data geocoding reaching an
acceptable planimetric positioning tolerance (depending on the
nominal scale of the base map that will be adopted and on the
final application required). Therefore MIVIS image geocoding
is a delicate step to go through; complexities are due both to the
whiskbroom MIVIS sensor model, which introduces many
deformations to take care of, and to the moved surface of the
area. Scene geometry has therefore to be corrected. Usual
procedure based on simple flat transformations cannot model
such geometry especially in a mountain region as the study area
is. Orthprojection has to be considered in order to make MIVIS
data suitable for the data integration and analysis.
Since MIVIS raw data often are released to the final user
without any metadata about attitude ad position time-dependent
of the sensor, a non parametric approach has to be applied. Our
first task has been to investigate which solution was the most
appropriate. We considered successively two orthoprojection
methods.
2. GEOMETRIC CORRECTION METHODS
2.1 Rational Function Model
This is the most famous and used non-parametric model. It is
present within almost every remote sensing commercial
software. It allows to relate image coordinates (5,77) with
object-terrain 3D coordinate (X,Y,Z) through rational
polynomials as shown in (1):
Spr YLT)
Fa 1:5] (1)
PX. Y,Z)
— P(XY,Z)
ET P
