International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B5. Istanbul 2004
YA
400m mb — 33.000
d——c——- -
200 m
100m _
Figure 3. Camera constellation
The image acquisition was carried out from the top of two high
buildings close to the groyne field. On each building two
cameras were set up. Due to the altitude of the camera positions
of about 40 m and a maximum distance of 400 m at the outer
boundary of the area under investigation, the cameras point
downwards with an angle of approximately 10 degrees. This
camera constellation results in two overlapping stereo models
I/II and I11/1V (see Figure 3). The orientation of the images was
established manually after image acquisition. The orientation
parameters are assumed to be constant for the acquisition of an
image sequence.
The centre of the area to be investigated has a distance from the
cameras of approximately 200 m. At this distance the images
have a scale of about 1 : 16 000. Assuming an accuracy of the
image coordinates of 1 pixel (6.7 um) in image space, a
theoretical standard deviation of 10.7 cm in X and Z, and
142.9 cm in Y is obtained using standard error propagation
formulae for the stereo pair I/II (see Figure 3). The poor
&ccuracy in Y reflects the small stereo base. From the point of
view of the application, however, the Z-accuracy is the most
critical one.
5. RESULTS AND ACCURACY CHECKS
For the generation of the dynamic DSM approximately 100
well distributed seed points were measured manually in the first
stereo pair of the image sequence I/II. Using these seed points
approximately 20 000 conjugate object points were determined
automatically by image matching. Subsequently, matching of a
5 min image sequence acquired with a frequency of 8 Hz was
carried out as described in section 2 and 3. The sequence
consists of 2 400 images.
In the following the image matching results are checked by
manual analysis and gauge data. First the comparison with the
manual measurement is carried out for one epoch and
afterwards with the gauge data for multiple epochs.
5.] One Epoch
Figure 4 shows the correlated points derived from the thirtieth
stereo pair of the image sequence I/II and the associated
orthophoto.
Figure 4. Correlated points and associated orthophoto
The image matching results were spot checked by manual
stereo analysis. The manual stereo measurements have been
carried out with the Image Station Z IV of Z/I Imaging. For
some stereo pairs the DSM has been acquired twice. The
standard deviation derived from double measurements amounts
to approximately 10 cm in the surf zone, the results in the
seawards area are slightly worse.
The manual measurement was undertaken for several stereo
pairs of the image sequence. At each time a part of the entire
matched area was analysed.
^ 1
A 41.0
i V^
/ \
[ A
1 i 2
i i fi 408
Y ii
V TNA £3
: VA [À 40.6
1 { —
\ = E
\ / x N
\ / 40.4
J
!
\ 17 .
Lf
À a
\ / 40.
V H
= image matching ;
manual measurement 4 0.0
A - À 1 1 - L A À
880 900 920 940 960 980
Y [m]
Figure 5. Comparison of image matching and manual
measurement (Profile through the DSM)
Figure 5 shows a comparison of image matching and a manual
measurement. The displayed data correspond to the profile
highlighted in black in Figure 4. The profile represents a cross-
section of the surf zone. Overall, the manual measurement is
much smoother than the matching results. The beach is located
at the left side of Figure 5. In the swash zone (Y = 892 m up to
913 m) slight differences between the manual measurement and
the surface derived by image matching arise. The small wave in
the surf zone (Y = 902 m) is well detected by image matching.
Discrepancies occur at the breaking wave (Y — 927 m) due to
the inexact determination at the sea spray. In the manual
measurement clearly identifiable points on the sea surface or on
the sea spray were captured. The matching algorithm, on the
other hand, compares the grey values of the left matching
Inte
—
win
righ
som
text
as v
mati
bou
altit
The
in tl
5.2
Noy
Thr
diff
part
ider
proc
sho
war
Dui
me:
met
Fro
sea
pro
Fig
one
56.
frec
the
The
inv
As
ma
the
anc
Als
ran
de
est
