International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B1. Istanbul 2004
4. DATASET DESCRIPTION
High resolution SPOT scenes captured by SPOT-1, SPOT-2,
and SPOT-S are used for evaluating the performance of the
proposed methodology for scene resampling according to
epipolar geometry as well as DEM generation. The following
subsections present a brief overview of these datasets as well as
some of the associated metadata.
4.1 Dataset 1
The first dataset involves a stereo-pair captured by SPOT-1 and
SPOT-2 over Korea, Figure 1. The specifications of these
scenes are listed in Table 1. As mentioned earlier, normalized
scene generation requires a minimum of five ground control
points to estimate the parallel projection parameters (Equation
2). For this dataset, a total of twenty-six ground control points
have been collected through a triangulation procedure involving
aerial imagery over the same area.
4.2 Dataset 2
A stereo-pair captured by SPOT-5 over Belgium, © CNES
(2003), distributed by Spot Image — all rights reserved, Figure 2,
has been used. The scenes have been provided by HRS
instruments as part of the Scientific Assessment Program (ASP)
to evaluate the quality of derived DEM data from SPOT-5. The
metadata associated with this dataset is provided in Table 2. A
total of nine ground control points have been collected to
estimate the parallel projection parameters. The ground
coordinates of these points are derived through a terrestrial GPS
survey by the University of Liege, Belgium.
p^
&
b) Right scene
a) Left scene
Figure 1: Original scenes for the first dataset
rap 5
a) Left scene b) Right scene
Figure 2: Original scenes for the second dataset
Scene Left Right
Satellite ID SPOT-5 (level 1A) | SPOT-5 (level 1A)
Focal length (mm) 1082 1082
Date of capture 2002-09-24 2002-09-24
Time of capture 10:53:07 10:54:39
Pixel size (mm) 0.0065 0.0065
3 of voi and 12000 x12000 12000 x12000
columns
Incidence angle (°) 23.031638 -22.471756
3round pixel (m) 5 5
Scene Left Right
Satellite ID SPOT-1 (level 1A) | SPOT-2 (level 1A)
Focal length (mm) 1082 1082
Date of capture 1998-10-29 1998-10-29
Time of capture 02:37:26.2 02:06:47.4
Pixel size (mm) 0.013 0.013
WOO s nd 6000 x6000 6000 x6000
columns
Incident angle (^) 20.2 -29.7
Ground pixel (m) 10 10
Table 1: Specifications of the first dataset
Table 2: Specifications of the second dataset
5. EXPERIMENTAL RESULTS
The proposed methodology starts by using a subset of the
provided control points to estimate the parallel projection
parameters in Equation 2. For the first dataset, six ground
control points are used for parameter estimation, while the
remaining twenty points are used as check points. Similarly, a
B total of six ground control points are used to derive the parallel
projection parameters for the second dataset, while the
8 remaining three points are used as check points. The square root
| of the resulting variance component (5) from the adjustment
procedure for the first and second datasets are listed in Tables 3
and 4, respectively. The variance component represents the
quality of fit of the observed scene and ground coordinates of
SE the involved points to the parallel projection model, as
described by Equations 2. Based on the numbers in Tables 3
and 4 (approximately one-third and four pixels for the first and
second dataset, respectively), one can conclude that the first
dataset shows a better fit. The poor fit associated with the
second dataset can be attributed to inferior quality of the
provided control and/or problems during the data acquisition
process.
Following the estimation of the parallel projection parameters,
the outlined methodology in section 2.2 has been used to
generate normalized scenes according to epipolar geometry.
The normalized scenes are generated to ensure the absence of
any y-parallax between conjugate points. The mean value of the
absolute y-parallax associated with the control and check points
for the first and second datasets are shown in Tables 3 and 4,
respectively. Closer inspection of these values verifies the
success of the proposed methodology in eliminating the y-
parallax between conjugate points. A stereo-anaglyph covering
the overlapping area in the second dataset is shown in Figure 3.
Three-dimensional visualization of this figure supports the
feasibility of using the parallel projection model for normalized
scene generation.
Int
Fit
pa
us
co
fir
re]
re:
"M Un - UD
gs és ar es