minimum period for collecting data is the time needed by the
incident wave to cross the surf zone from the seaward boundary
to the line of highest wave runup. For this process a duration of
about one minute is required (Niemeyer, 1997). Also, the
measurements have to be carried out over a period, which is
long enough to allow the analysis of the wave crossing and
runup. To obtain an acceptable standard deviation of the wave
parameters, some 200 waves have to be analysed.
Consequently, the chosen measurement system must be able to
sample data over a period of up to approximately 20 minutes.
In order to be able to describe the kinematics of the sea surface
induced by sea state a high resolution in time must be chosen.
The mean wave period in a groyne field near Norderney is
approximately 6 s. A sufficient frequency lies between 10 and
20 Hz.
3. IMAGING SYSTEM AND DATA AQUISITION
The measurements at Norderney Island are carried out from the
top of high buildings. The cameras II and III are set up in the
normal case (see Figure 2). To enlarge the base-to-height ratio
and thus to improve the accuracy the convergently arranged
cameras I and IV are added. At the same time the recorded area
can be increased by successively adding cameras in X-direction.
For data acquisition of the test area digital video cameras with a
2/3 inch interline progressive scan CCD are used. The CCD-
sensor has a radiometric resolution of 10 bit greyscale
(monochrome) and a geometric resolution of 6.7 x 6.7 um? per
pixel. The sensor size is 1300 x 1030 pixel, the maximum frame
rate is 12 frames per second. The system allows a maximum
observation period of approximately 20 minutes, the bottle-neck
are current disk limitations. The exposure time can be controlled
by an external trigger signal. For camera synchronisation IPI
developed a wireless system to transmit an external trigger
signal from a master station to all slave stations (three in this
case) approximately every 1.5 ms.
Ap — 10.000
Figure 2. Camera constellation in planimetry
In principle the achievable accuracy is influenced by the object
size, the number of available cameras, the focal length and the
camera locations. Using a focal length of 50 mm an area of
200 by 200 m? can be recorded with an accuracy « 4 cm in X-
and Z-direction and «8 cm in Y-direction at the seaward
boundary, assuming an image scale of 1:10000 and an
accuracy of the image coordinate measurements of 3.5 um,
corresponding to 0.5 pixel.
The orientation of the images is established manually after data
acquisition. The orientation parameters are assumed to be
constant for the acquisition of one image sequence.
4. MATCHING METHOD
The 3D recording of the wave surface from images requires the
interior and exterior orientation of the images and conjugate
points. Therefore, one of the major tasks during the
photogrammetric object reconstruction is the search of
conjugate points.
Since a number of years automatic matching methods have been
investigated as a major issue in digital photogrammetry. The
automatic methods for image matching can be divided into three
classes, area-based, feature-based and symbolic or relational
matching (Schenk, 1999). In area-based matching entities are
the grey levels of small areas of two or more images and
matching is carried out by cross-correlation or the highly
accurate least squares technique. The latter method requires
very good initial positions. Feature-based matching determines
the correspondence between points, edges (e.g. the wave runup
line) or other features derived from the original images. The
similarity (e.g. the shape, sign and strength of the runup line) is
defined based on a cost-function. The symbolic matching
method refers to methods that compare symbolic descriptions of
images, for example the breaking waves, and measures the
similarity also by a cost function.
These matching methods exist and work well for many
photogrammetric applications. Examples for the matching of
sea surfaces are given in (Redweik, 1993), (Taguchi, Tsuru,
1998) and (Yamazaki et al., 1998). However, the authors are not
aware of any software, which is optimised for the matching of
wave surfaces. First tests have been carried out with IPI's
matching software DPCOR. This software has been used in a
large number of photogrammetric projects before, e.g. (Heipke
et al, 1994), (Heipke et al., 1996), (Rieke-Zapp et al., 2001).
Conjugate seed points have to be determined manually, then the
algorithm follows the region growing principle to match
conjugate points in stereo image pairs (Otto, Chau, 1989).
Matching is carried out in image space, no orientation
parameters are required.
5. IMAGE SEQUENCE MATCHING
During the analysis of image sequences the predictable motion
of surface models can be used as additional information. The
established temporal correspondence between the subsequent
frames of an image sequence can in particular increase the
reliability of the results.
The image sequences are obtained with a frequency of 12 Hz.
Thus the basic idea of the used processing principle is that
subsequent images of one sequence do not change very much.
So it should be possible to feed the matching programme with
manually measured seed points of just one or a few stereo pairs
at the beginning of a sequence, and the program should find the
needed seed points of the following stereo pairs on its own. In
the following, such an approach is described. First the
movement between the subsequent images is assumed to be
zero. In an enhancement of this procedure the motion of the
wave surface is taken into account (Kónnecke, 2002).
—597—