pt for
ciety.
basic
vith a
jects,
tions,
y for
hical
duct,
n the
thods
neral
ensor
uring
areas,
asily
ation
matic
ectral
i med.
ends
for
' two
1ding
s, an
> line
lized
f the
nakes
lems
| the
'S are
n the
e the
rector
, itis
easy to detect and minimize these differences by a gain/bias
replacement in raw images.
SPOT imagery, geometrically uncorrected, often displays a
regular chess-pattern noise with a period of two pixels. This
problem occurred in the early life of each satellite, but has since
then been reduced. The reason of the noise is still not fully
understood.
A filter is derived, and used by SSC Satellitbild, to remove this
coherent noise from the un-resampled image before the
geometrical correction. (Westin, 1990)
In areas with low reflection i.e. forests and areas with low sun
elevation relevant information is registered in only 10-20
quantum levels. The signal to noise ratio is reduced which
causes stripings in the images using traditional contrast
stretching methods. The problem is solved by multiplying the
raw image with a suitable factor before applying the calibration
coefficients during the contrast stretch.
2.3 Geometric correction
Due to the stable orbit and the fact that satellite motion is
subject to the laws of celestial mechanics, it is possible to
formulate a mathematical orbital model describing the exterior
orientation with a reduced number of parameters. The
orientation of one line could then be evaluated based on
knowledge of the time for the line registration.
23.1 Orbital Model. The satellite image includes a two-
dimensional coordinate (line, pixel), which has to be correlated
to a three-dimensional geodetic coordinate (latitude, longitude,
height). This is done by measuring Ground Control Points
(GCPs) i.e. with GPS or by digitizing objects from maps, and
locating them in the image. These earth observations are used to
update the a priori values of the parameters in the orbital
model, and this makes it possible to correlate and resample
satellite registered information in accordance with the ground
truth.
The following list is one way of describing the complexity of
the mathematic orbit adjustment model. Transformations
between these coordinate systems are calculated, beginning
with the information registered by the sensors in the satellite
and ending with the ground control points.
* The Sensor Coordinate System describing the detector
position errors in the CCD arrays.
* The Attitude Measurement Reference System includes
information of the discrepancies in attitude angles
according to the zero a priori value.
* The Local Orbital Reference System is a moving system
describing the position of the satellite mass ceníre in
reference to the earth centre.
* The Earth Centered Inertial Coordinate System (ECI) which
has its origin at the mass centre of the earth, and also the
orbital parameters referred to it.
* The SPOT Ephemeris Reference System uses the
International 1980 ellipsoid for the ephemeris. The
information has to be transferred from the SPOT system to
the ECI system before it can be used for calculation of
orbital a priori parameters.
* The Ground Control Point Reference System describes the
local geodetic system in which the ground control points
are measured.
23.2 Sequence of scenes. The continuous registration,
"Pushbroom scanning", used by the SPOT satellites makes it
possible to extend the geometrical correction of one scene to
Include scenes registered in sequence. The adjustment model
Will be more complex but it results in a stable adjustment that
could include up to five scenes. This method decreases the
Production time and reduces problems of merging scenes
registered in the same strip. (Westin, 1991)
23.3 Ground Control Points (GCP). Topographic maps are
usually sufficiently accurate for the correction of satellite
Images on scales around 1:50,000. Investigations show that
there is no particular increase of accuracy, with more than seven
P's per satellite scene. Experiences from precision
correction using maps on a scale of 1:50,000 show RMS errors
of approximately 20 metres, caused by digitizing errors, map
shrinking and inaccurate geographical position of the objects.
Instead of using maps with low or unknown accuracy, it is
possible to produce satellite images as a concept for the control
point measurement. By using three to four points within each
raw scene it is possible to optimize the GPS measurement
according to the orbital model and geometrical correction of a
sequence of scenes, still achieving RMS errors during the
adjustment of around five metres.
The optimal method of measuring GCP’s with Global
Positioning System (GPS) was investigated and evaluated by
students from the Royal Institute of Technology in Stockholm,
(Dahlgren, Svedjesten, 1993). Two methods briefly described
as follows was evaluated according to the criterion to find out
methods resulting in RMS errors not exceeding five metres.
Static observation, occupies the control point until enough data
has been registered from the satellites. This is the most accurate
method and the one traditionally used, resulted in a RMS of 3.4
m during the geometric correction in 2-dimension.
The results of the investigation show that an increase of the two
parameters of major interest for the evaluation, observation time
15 min and baseline distance 50 km, did not significantly
improve the insufficient RMS error.
Kinematic observations, allow placement of the receiver on the
roof of a car. Measurements can then be performed during the
movement through the road intersection. This measurement was
done from both the crossing roads and results in two crossing
vectors, which allows calculation of the centre of the
intersection. The result of the observations, RMS 4 m, shows
that this technique has a very high potential for automatic
processing in the nearby future.
2.3.4 Digital Terrain Model (DTM). It is necessary to use a
Digital Terrain Model, DTM, to reduce the influence of the
oblique viewing angle and the terrain altitude. The DTM could
be processed from a SPOT stereo pair by automatic matching,
using methods as Multi-Point-Matching (Rosenholm D., 1987)
or Match-T. For reasons of cost and time it would not be
rational in these projects since contour lines already exist in the
old maps.
Digitized contour lines from different map sheets was edge
connected to each other before any further processes could take
place. The digitized and coded contours was then transformed
to the Baltic system, using transformation parameters, described
in paragraph 2.3.7.
The digital contour lines was used as input to a Digital Terrain
Model interpolation program TIN which creates triangles
according to the digitized breakpoints in the vectors. The
regular DTM grid was then interpolated from the three height
values of the triangles. It is important to use approximately the
same distance between the breakpoints in one contour line as to
the nearest digitized line with different elevation.
Satellite GPS- Digital
Images measurments Terrain Model
Orbital Satellite
Adjustment a priori data
Ortho
Image
Fig. 1. Geometric correction
455
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996