International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part BI. Istanbul 2004
2.4 Digital surface model and slope and aspect models
The HRSC-A images were processed automatically by DLR
(German Aerospace Center) in the photogrammetric processing
described by Wewel et al. (2000). This process permits
production of a digital surface model (DSM) of the imaged area.
Two raster models for both test sites were produced, one per
flight session, with a one-metre cell size. The height resolution
of these models was five centimetres. The slope and aspect
raster images were calculated to characterise the surface
slanting and derive a surface normal for each pixel. The spatial
resolution of these models was one metre.
2.5 Sun angle data
The sun angles at the time of image acquisition were determined
using time averages for each image strip. A feasibility point
near the crossings of the flight strips at both test sites was used.
At the Sjókulla test site, the feasibility point was located at 60?
14' 31.07"N, 24? 23' 3.00" E. In Kuckuberg, the corresponding
point was located at 60° 12° 30°°N, 24° 27’ 00" E.
Sun angles were calculated at one-minute intervals. Angles for
the full minute nearest to the average time of an image strip
were chosen. The rounding causes some error in the sun
direction at the ends of each image strip, but due to the short
image acquisition period of a single strip and the even smaller
intensive test area this does not have a significant effect. The
estimated azimuth error was 0.30° and the zenith angle error
0.06° at the ends of each image strip.
2.6 Ground data
The test sites were examined for ground truth data at the time of
the HRSC-A flights. Documentation of the scenes was
performed by photographing the targets, determining the species
and defining the areas on to a map. Vegetation and the current
state of growth were also recorded.
For sensor calibration, three additional known reference targets
were spread over the area of the photogrammetric test field,
which has dark gabro gravel as background. The reflectance
properties of the reference targets have been determined by
laboratory calibration tests and they will be used for sensor
calibration.
2.7 Sample point determination
To derive a proper reflectance sample with correct angular
measures, the image acquisition had to be reconstructed. Using
the position and attitude data of the camera and single optics
camera model, any image cell on a CCD line sensor could be
captured with the known viewing azimuth and zenith angles
and the corresponding target point on the ground surface.
»
An HRSC-A camera has nine parallel CCD line sensors
mounted with specific measuring angles with an 11.8° swath
and the location of a sensor cell could be determined using these
angles. The height reference surface could be reached by means
of a rotation of the constructed image vector with a 3D rotation
matrix and by scaling the result
Mox
^
=MR.R,R,|v|, (2)
K^ Nw
N
N
where x, y and z are sensor cell coordinates on the image plane
and the triplet X, Y, Z is a vector from the projection centre of
the camera to the ground. M is the scale factor.
The actual sample point (measured by the camera) was
determined by finding the intersection of the target vector and
the digital surface model. This was carried out using a bisecting
algorithm. The search area for the intersection point in the
direction defined by the direction vector was initialised to begin
from the maximum height of the surface model and to end at the
minimum height of the surface model.
The intersection point lies between these two preset end points.
The search for the intersection point was carried out bisecting
the search area, or vector, and checking whether the middle
point is above or under the surface. If it is above the surface, the
beginning of the new search area is set. Otherwise, it is set to be
the end of the search area. Iteration continues until the distance
between the height surface and iterated point is less than a
preset threshold.
The threshold value used in this study was five centimetres,
which corresponds to the height resolution of the digital surface
models. For a maximum HRSC-A view zenith angle of 18.9°,
five centimetres threshold means less than two centimetres
uncertainty on the horizontal ground plane, which is acceptable
for even more detailed image resolution.
The 11.8° opening angle of a single line sensor was divided into
200 parts, giving an angular separation of 0.059° for adjacent
samples. This corresponds to a distance of approximately 3
metres between the sampled data points in a cross-track (i.e.
sensor) direction from a flight altitude of 3000 metres. In flight
direction, the sampling interval was approximately 3.5 metres at
flight a speed of 69 m/s and was attained by linearly
interpolating the sparse one second interval GPS data and 1/10
second interval inertia data to a 1/100 second interval dataset
and picking every fifth point with its position and attitude data
from the set. Linear densification is considered to have only a
minor effect on data quality.
2.8 Geometric measures of BRDF: flat and slanting surfaces
The BRDF geometry was first investigated for a flat surface.
The view zenith angle of a sample was calculated simply, using
the target and zenith vectors. The azimuth of the target vector
was measured clockwise from the north. Sun angles were
calculated as described above in caption 2.5 and were applied
here. The relative azimuth of the viewing and illumination
directions was calculated.
Regarding each DSM point as a slanting plane, the BRDF
coordinates could be determined in relation to the normal vector
of this plane. The aspect value was corrected for meridian
convergence and the normal » was constructed out of the
corresponding slope and aspect values of the sampled object
point. Sun and viewing vectors § and ¢ were constructed
using the corresponding zenith and azimuth values (64, 0,9 and
OQ, $5) Viewing and illumination zenith angles @. and 6,
relative to the surface normal were then calculated.
Intei
—
The
dete
proj
uniq
plan
Wri
becc
whe
rela
arbi
2.9
Atır
obst
oi .
diff
con
the
gro,
of
Atn
anal
The
esti
ther
met
Thi:
patl
one
radi
spe
doc
met
cor
