of 10 W halogen lamps which can be easily attached to
the net by cable clips.
The whole pen structure consists of different parts,
which can move slightly with respect to each other while
the pen is floating in the water. It is impossible to have
totally fixed control points on the object. Some points
(lights) are attached to vertical steel pipes which are fixed
to the pen structure. During the photogrammetric image
capturing nobody is allowed to be on the pen so that the
movement will be reduced to a minimum and these
points can be used as fixed points. Distances between
the fixed points are measured to obtain a correct scale.
Distances from the fixed points to the water surface are
measured to level the coordinate system. The x- and y
directions of the coordinate system are chosen arbitrarily.
2.2 Hardware System
The underwater imaging system must provide high reso-
lution and sensitivity. Due to the fact that the target
points are not labelled, a continuous video is appropriate
to track the points through the different images.
The system is composed of the following components:
e Underwater CCD video camera with good low-light
sensitivity (Micro SeaCam 1050, focal length 2.8 mm
(air), fixed focus, wide angle, depth of field 10 cm to
infinity, 1/3 inch CCD sensor (537x595 pixels), 8bit
grayscale).
e Cable to carry power and video signals between cam-
era and surface (2 x 40 m).
e S-VHS Video recorder unit (Panasonic AG-4700E).
e Video frame-grabber (Cameron CamDrive, PAL Video
Signal, 800x600 pixels, PCMCIA).
e Portable computer (Design-Notebook, 486DX4-100).
2.3 Camera Calibration
The focal length of the camera in water is different to the
focal length in air because of the different optical densi-
ties. The optical density varies due to the influence of
salinity, temperature and water depth. Therefore an on-
site camera calibration in seawater has to be performed.
There are parameters for linear (scale difference as and
non-orthogonality a4) and radial distortion (a1 and az) as
well as the principal point (xo and yo) and the focal length
(c) (Krzystek, 1995).
A flat test field with a large number of points (cf. Figure
1) is used for the calibration. The point measurement can
be done automatically by least squares template match-
ing for most of the points. Some are disturbed by reflec-
tions and had to be measured manually. In the following
bundle adjustment the calibration parameters are ap-
plied. The calibration results are shown in Table 1; the
distortion vectors are shown in Figure 2.
The calibration results lead to corrected image coordi-
nates with an accuracy of 3 um (1/3 pixel).
526
2.4 Procedure
Depending on the pen setup, the recording device is
operated from the pen platform or from a boat. The cam-
era is connected to the recorder by a 40 m or 80 m video
and power cable. Since we used an off-the-shelf video
recorder without special housing etc. this equipment has
to be protected from sea water and bad weather condi-
tions.
The camera is operated by a diver who swims around the
pen at different depths and distances, panning the cam-
era left and right as well as up and down. He cannot see
the image contents since there is no viewer at the cam-
era. The recording has to be supervised from above,
where it can be checked using a TV-monitor. This leads
to rather long recording times and should be improved.
The recorded tapes are reviewed and appropriate images
are selected by the opeator. The images are transferred
to the notebook hard disk as 800x600 pixel TIFF-Files
(cf. Figure 3) using the PCMCIA frame grabber. With
these images the standard photogrammetric procedures
(interior orientation, point measurement, bundle adjust-
ment) are performed.
2.5 Digital Photogrammetric System
The photogrammetric aspects of the CoSMoLUP system
are based on the PICTRAN system's modules for digital
photogrammetry. Both systems run on standard PCs
under Windows 3.1.
For CCD-cameras the interior orientation is constant; so
the transformation from the pixel coordinate system to
the image coordinate system can be written to the im-
ages directly without using any fiducial marks.
The PICTRAN modules offer the opportunity to measure
image coordinates of signalised points manually or by
using least squares template matching. For most of the
points the semi-automatic measurement could be used,
others had to be measured manually because of poor
visibility (contrast and size).
The integrated bundle adjustment does not need any
approximate values for projection centres, image rota-
tions or object coordinates. In a first step the images are
oriented by using only a few of the points which can be
identified properly (in general the fixed points). Since the
points all look very similar and there is no background
information in the images it is sometimes very difficult to
identify the points. In the second step this preliminary
orientation information is used to display the epipolar
lines of the measured image points in all of the other
images. This helps to identify the same object points in
the different images (cf. Figure 5). After having measured
all image points the bundle adjustment was performed
again and the object coordinates of all points determined.
2.6 Results
In order to check the internal accuracy measures, two
different operators determined the object coordinates
independently - from selecting the images to the object
coordinate determination. In this test three different pen
types and sizes where measured, but only with a mini-
mum number of points on two sides of the nets (not
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B5. Vienna 1996
sufficie
Figure
net 2 is
circular
of net '
could i
other n
not as
In all t
compa
and m:
The di
the rar
result
ences
racy.
The ca
each c
analys
taken
improv
The re
and pe
measu
earlier