International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B1. Istanbul 2004
3. DATA EVALUATION
The work flow can be subdivided into 7 steps:
l. Direct geo-referencing of the images using the supplied
orientation parameters (auxiliary data)
2. Measurement of control and check points in the 8 TS
3. Mass point generation in image space using region
growing matching algorithm.
4. Bundle adjustment estimating correction polynomials
5. Transformation of image mass points into object space
using both strict model and rational functions
6. DSM generation using both types of data sets
7. Comparison of the DSMs with the reference DTM and
statistical evaluation.
3.1 Direct Geo-referencing
In a primary step the images are directly geo-referenced
applying the model described in the SPOT Satellite Geometry
Handbook (SPOT Image, 2002) using the supplied orientation
parameters and look angles. The main purpose of this step is to
check the consistency and the quality of the auxiliary data.
Comparing the resulting horizontal coordinates and the
reference coordinates deduced from the orthoimages for a
measured sample point, differences of less than 30 m are
obtained which impressively demonstrates the high quality of
the supplied SPOT-5 auxiliary data. The program later is
extended to consider the estimated correction polynomials (see
3.4.1) and is used for the calculation of the cube data, needed
for the computation of the rational functions (see 3.6)
3.2 Measurement of control and check points
It was planned to have a group of 5 points for each test site to
serve either as control or as check points. They are measured
by an experienced operator on a digital photogrammetric
workstation with matching support in the HRS and THR
images and also in the respective orthoimages. For time
reasons only one of the four orthoimages per testsite is used.
For a better point identification the different image scales of
the HRS images are adapted applying a scale factor of 2 in
scan direction. Nevertheless, the point identification resulted
to be very difficult, especially in the scaled oblique looking
HRS images. Only points lying on the ground surface can be
selected, since the reference heights are taken from a terrain
model. Due to the presence of forests and buildings in wide
areas of the test sites #3, #4 and #6, there it was not possible
to measure all 5 points. 19 points of the TS #1, #3, #7 and #8
are taken as control points and 17 points of the remaining TS
#2, #4, #5 and #6 as check points (see figure 3).
3.3 Mass point generation in image space
For the mass point generation in image space a modified
region growing algorithm, originally developed by (Otto and
Chau, 1989), is used, which already has successfully been
applied to SPOT-1 images in the early 90es (Heipke and
Kornus, 1991). Starting from a couple of manually measured
so-called seed points, the algorithm matches the four
neighbour pixels (left, right, upper and lower) at a given
distance. For this study 1 pixel distance of the original HRS
image is chosen. If the matching result meets some specified
criteria (a minimum correlation coefficient, a maximum
number of iterations, etc.) the point is added to a list and
serves itself as a new seed point. The process ends after all
points of the list are matched and no more neighbours can be
found, which meet the criteria.
The algorithm is applied to sections within the 8 test site of
the HRS1, HRS2 and the THR images with a size of approx.
2700 x 3300 THR pixels. For each test site 3 matching
combinations are calculated: a) THR — HRSI, b) THR -
HRS2 and c) HRS2 — HRSI. The THR-points successfully
matched in the 1* combination are entered into the matching
of the 2" combination as so-called transfer points, i.e. only
the coordinates in the second image are determined while the
coordinates in the first image resulting from the previous
matching run are kept. Accordingly, the resulting HRS2-
points of the 2'* combination again form the transfer points
of the 3™ combination. The points obtained from the 3
matching runs are classified into 3 groups, depending on the
number of combination they have been matched:
1. in 1 combination (THR-HRS1 or THR-HRS2 only),
2. in 2 combinations (THR-HRS1 and THR-HRS2 or
THR-HRS2 and HRS2-HRS1) and
3. inall 3 combinations.
Table 1 gives a survey of the matching results achieved in
the 8 test sites. While group 1 only contains 2-ray-points,
group 2 contains 3-ray-points and group 3 redundant 3-ray-
points with HRS1 coordinates matched in two different
matching runs. The two corresponding coordinates are
averaged and deviations from the average AX?! and AY! are
used to calculate standard deviations Gax , Oay°!. For the
later DSM generation only 3-ray-points were taken with a
correlation coefficient p bigger than 0.7 and, in case of point
group 3, with deviations AXS!, AYS! smaller than three times
their standard deviations Gax°!, Gay”.
From point group 3 also a subset of points is selected as input
for the bundle adjustment using a regular grid of 100 x 100
pixel mesh size. Taking the point with the maximum
correlation coefficient within a grid mesh, 5267 regularly
distributed tie points are obtained within 7 TS (see figure 3).
From TS #2 no tie points are extracted in order to serve as a
real check area.
Number of TS 1 2 3 4 5 6 7 8
2-ray-points 75643 — 152505... 71656... 79575 1196941 72379. „54551. 100053
2-ray-points with p >0.7 39108 134013 29233 57965 103744 49391 41459 88808
3-ray-points matched in 2 combinations 166657 205687 177825 170871 109805 136127 143485 137034
3-ray-points in 2 combinations with p > 0.7 85497 164198 58468 113097 87566 95301 82035 117369
Points matched in 3 combinations 728825 678645 467015 596486 768396 541319 797748 686262
Points in 3 combinations with AS < 35,5! 595051 601042 393743 485450 678165 454615 697439 595961
Total 3-ray-points selected for DSM generation 680548 765240 452211 598547 765731 549916 779474 713330
Table 1: Results of region growing image matching
446
Intei
TOW
3.4
In ©
eithe
data
impl
(Col
The
give
cam‘
cam
3.4.1
The
the |
to t
defir
Yo, 4
