inbul 2004
ACCURACY ANALYSIS AND SURFACE MAPPING USING SPOT 5 STEREO DATA
Hannes Raggam
Joanneum Research, Institute of Digital Image Processing
Wastiangasse 6, A-8010 Graz, Austria
hannes.raggam@joanneum.at
Commission I, WG I/2
Key Words: Spot 5, Sensor Model, Accuracy Analysis, Image Matching, Elevation Model
ABSTRACT
On the Spot 5 satellite stereo data can be acquired simultaneously from the HRS (high resolution stereoscopic) instrument, which
comprises 2 cameras looking forward and backward, respectively, at an off-nadir angle of +20 degrees. Enhanced along-track pixel
resolution of 5 meters shall further assure a high accuracy with respect to 3D data extraction. In this concern, an HRS study team
was installed and specific test sites designated in order to validate the geometric performance of Spot 5 HRS stereo data. This paper
summarizes the results which have been achieved from the validation activities of the Institute of Digital Image Processing. The
investigations were carried out using a test data set acquired over the city of Barcelona. Study areas showing different topographic
characteristics have been investigated. Beside pure HRS stereo data, a glimpse has been further made onto the joint use of the HRS
stereo data and a supermode THR image, which is basically a nadir scene with a pixel size of 2.5 meters. |
1. INTRODUCTION
In order to continue and strengthen the traditional Spot
stereoscopic potential, the HRS instrument was realized on Spot
5. allowing the acquisition of in-track stereo data during the
same overflight. The HRS instrument comprises two CCD
cameras looking forward and backward, respectively, at an off-
nadir angle of +20 degrees, allowing stereo data capture at a
time interval of about 90 seconds (Gleyzes et al., 2003). The
resolution of the stereo data is 5 meters along and 10 meters
across-track. Thus, the performance of image matching applied
to such data is not degraded due to temporal changes in the
scenes.
An HRS assessment program has been established, involving an
expert team as well as a number of designated test sites in order
to verify mapping accuracies being predicted for Spot 5 HRS
stereo data. Following the objectives of this program, this paper
is devoted to the analysis of the topographic mapping potential
of Spot 5 HRS stereo data. The following issues are specifically
addressed:
e Geometric modelling of HRS stereo data, including
utilization of given orbital and imaging parameters
e Analysis of a-priori 2-dimensional location accuracy for
individual scenes as well as 3-dimensional location
accuracy of stereo models using control points
e Optimisation of sensor models utilizing control point
measurements and least squares parameter adjustment
e Discussion of the performance of image matching
techniques being applied to Spot HRS as well as multi-
sensor THR/HRS stereo data
e Extraction of surface models from HRS as well as multi-
sensor THR/HRS stereo data
e Accuracy analysis of extracted elevation models through
comparison with reference data
For the study the Remote sensing Software package Graz (RSG,
Joanneum Research, 2003) was used. This is designed for
geometric processing of remote sensing images, including the
aspects of geometric sensor modelling, block adjustment, stereo
image matching, and surface model generation. At the Institute,
Spot stereo mapping using analogue Spot stereo pairs and an
analytical plotter was already an issue in the eighties (Raggam
et al, 1989). Also the aspects of multi-sensor stereo data
utilization as well as validation of stereo mapping potential
(Raggam et al., 1990 and 1996) have been a matter of previous
research.
2. TEST DATA
The city of Barcelona as well as the areas north(-west) of it
were selected as a primary test site for the HRS assessment
program. The related test data set was provided for this study
and comprises the following image and reference data:
e a Spot HRS stereo pair with a nominal pixel size of 5
meters along- and 10 meters across-track;
e a THR supermode product with a nominal pixel size of
2.5 meters;
e a digital elevation model with a mesh width of 15 meters;
e ortho photo mosaics for 8 detailed study areas with a pixel
size of 0.5 meters.
Some 15 high precision control points were measured for each
of the detailed study areas from the respective ortho photo
mosaics, resulting in an overall pool of 116 points to be used
for sensor modelling and accuracy analyses.
3. SENSOR MODÉLLING
In general, rather detailed and precise information is provided
for the sensor model of a Spot 5 image, comprising orbital data
as well as attitude parameters of the instrument (Spot Image,
2002). This justifies the feasibility to generate value-added
products: with high accuracy. Geolocation accuracies in the
range of 15 meters in planimetry and 10 meters in height are
anticipated with respect to HRS input images without using
control points for optimisation (Spot Magazine, 2003).
