{RS data
est a new
1e model
nt during
nodel the
it of each
rigorous
EM of a
- satellite
| DIMAP
1cts.
E France
n Figure
/ Terrain
umber of
types of
— 4344!
— 4334
TKIN
43°12
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part Bl. Istanbul 2004
The ground control points were originally provided by IGN
for the OEEPE test of SPOT data and were mainly extracted
from 1:25000 maps. A total of 38 GCPs were measured in
HRS images having a very good distribution all over the
whole HRS images. As will be mentioned later, the
distribution of the GCPs all over the images is very important.
4. SENSOR MODEL
4.1 Application of the sensor model
4.1.1. Fundamentals. For this assessment various versions of
the UCL satellite sensor model are used, depending mainly on
the reference coordinate system. All versions rely on the
rigorous collinearity equations with the following
assumptions:
+ The satellite is moving along a well defined, smooth, close
to circular elliptical orbit.
* The images are acquired with a pushbroom scanner using a
constant time interval. As a result the coordinates along the
flight path have the same scale.
* A single image consists of a number of consecutive one-
dimensional scan lines. The relationship among sampling
lines is characterized by the dynamic orientation
parameters which are modelled with low order
polynomials as a function of the sampling time.
+ A stationary world is assumed, and a moving camera.
* The sensor array is approximately perpendicular to the
direction of motion.
* Attention must still be paid to the solution instability that
can arise from over-parameterization of the model. .
»* The attitude (o, ¢ and k rotations) of the satellite remains
constant during the acquisition time of each image with
respect to earth reference coordinate system.
* During the satellite's flight a perspective projection is
maintained across track. On the other hand a curvilinear
projection is maintained along the flight direction.
A specific model for along track stereo satellites sensors
using photogrammetry in combination with astrodynamics is
also used in an Inertial Coordinate System. The fundamental
point of this model is that, the motion of the satellite is a
Keplerian motion during the acquisition of along track stereo
images. For HRS stereo images the acquisition time interval
between the two images is about 91sec. Generally, for all the
models the position the velocity vector and the rotation angles
are computed with respect to Reference Coordinate system.
4.1.2 Flexibility of the sensor models. The developed sensor
model is very flexible. It is possible to have the following
solutions:
* à direct solution using the information provided from the
metadata file;
+ a refinement of the direct solution using one or two GCPs,
including also the self calibration process;
- using three or more GCPs without use any information for
exterior orientation from the metadata file.
On the other hand, the specific model for along track stereo
satellites sensors will be improved so that the information
which is extracted from the metadata will be used not only to
solve the model, but also, to refine and improve the solution.
Finally another model for along track images has already been
developed, where the orbital elements are computed directly
using three GCPs. The along track images are treated as one
entity, where the unknown parameters are the orbital elements
of each image instead of the state vector.
4.1.3 Image space coordinate system. For push broom data
one sampling line can act as the base line for computing the
exterior orientation parameters of others lines; this line is
assumed to be the centre line of the pushbroom image, and the
origin is the middle point of this line, because the acquisition
time of this line is known, accurately. The directions of the
image coordinate system are the following:
* The x-direction is the flight direction.
* The y-direction is perpendicular to x-direction
4.1.4 Object space coordinate system. In this assessment,
mainly a geocentric coordinate system is used for the sensor
model testing. The position and the velocity vector as they
extracted from the metadata file are in the ITRF90 geocentric
reference coordinate system. Because the WGS84 is very
close to ITRF90 with accuracy better than a meter, the
WGS84 1s used as the default geocentric coordinate system.
Also the sensor model is solved in the default geodetic system
of the area of interest which is the French NTF system. The
reference ellipsoid is Clarke 1880. The Lambert III projection
is used.
Finally an inertial coordinate system is used, in order to meet
the principal assumption of Keplerian motion of the modified
sensor for along track stereo satellite sensors.
4.2 Calculations from the metadata file
The following information from the metadata file is used in
the sensor model in order to solve directly in WGS84:
= Position and velocity vectors of the satellite measured by
the DORIS system every 30 seconds with respect to
ITRF90 (International Reference Frame 1990).
= Absolute attitude data measured by the on-board star
tracking unit for about seven times per second with respect
to the local orbital coordinate frame.
= The look direction table for the central pixel of the array.
= The scene centre time and the sampling time.
For the direct georeference procedure some calculation should
be made in order to find the position vector, the velocity
vector and the attitude in the centre of each image.
4.3 Solution in Geocentric Coordinate system (WGS84)
4.3.1 Introduction. As already mentioned, the sensor model
could be solved directly using the metadata information
without ground control points. Unfortunately, because the
rotation angles of the centre point of HRS images are not
calculated correctly, the rotations are treated as unknown
parameters. In this case two GCPs are needed for the solution
where the state vector has already been extracted from the
metadata, correctly.
4.3.2 Direct solution. The next step is to check the stability
and the rigorousness of the model itself, because the position
and the velocity vector are calculated from the metadata not
from the resection process. Seventeen independent check
points (ICP) are selected covering the whole area of the
images. The same points are used within the tests. The main
point is to understand the behaviour of the solution. Using 17
check points the intersection is solved and the results in
WGS84 geocentric coordinates given in table 2.
Having in mind what is mentioned in the section 2.2, that the
planimetric accuracy should be between 10-15 m the achieved
accuracy is in within these limits. Although the following
comments should be made:
