The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B5. Beijing 2008 
738 
location of sampling on the seabed is critical to relate the 
sampled area to environmental co-variates extracted from 
hydro-acoustic and other sensors. Calibration and data 
processing requirements for the various sensors is described in 
Williams et al. (2007) and Kloser et al. (2007). 
A separate forward-looking camera provides an additional view 
for navigation and obstacle avoidance. Additional sensors 
record altitude, pressure, pitch, roll, water temperature, 
conductivity and fluorescence. All sensor data is captured to a 
log file and combined with vessel DGPS and USBL information. 
Several sources of incoming data are displayed graphically on a 
custom-made Lab View “console” on an onboard PC screen to 
provide feedback to the pilots for control of the system. The 
console is also the switching interface for components. AC 
power is supplied to the system from the ship. Two 250 watt 
incandescent lights provide illumination for the video cameras. 
Strobes provide illumination for the digital still imagery. 
Figure 2. CSIRO towed body platform 
The towed body is deployed over the stem of the vessel using a 
gantry and is towed at an optimum speed of 1-1.5 knots that 
enables the pilot to “fly” the platform just above bottom. The 
cameras view the sea floor obliquely from 1-3 metres above the 
seabed. Deployments are typically 30-60 minutes duration, 
producing transects of 1-3 km in length, but, if required, the 
body can be towed continuously for several hours. 
The resolution of the video images is a limitation, so high 
resolution digital still images enable qualitative analysis at a 
greater level of detail. The digital still camera is remotely 
triggered by the operator or programmed to fire at set intervals. 
Images are captured to the internal storage of the camera and 
later uploaded to the logging computer. At this stage there are 
no plans to incorporate a pair of digital still cameras, although 
stereo digital stills have been used very successfully for some 
under-water applications of quantitative measurement (Abdo et 
al, 2006). However as a measure to overcome the limitation of 
PAL video resolution, high-resolution (1392 x 1040 pixel) 
progressive scan cameras are under evaluation for the stereo 
video imaging, based on experience with a proto-type system 
used in aquaculture (Harvey et al., 2004). The high resolution 
images improve the measurement accuracy from the stereo 
image pairs, the cameras are accurately synchronized and image 
sequences are recorded direct-to-disk in readiness for 
immediate analysis. 
3. STEREO-CAMERA CALIBRATION 
3.1 Shallow Water Calibration 
Video cameras used for marine science applications are not 
purpose-built for accurate and reliable measurements from the 
captured images, but instead follow different design imperatives 
to optimise the quality of the images and the utility of operation. 
Underwater use introduces another level of complexity because 
of the additional effects of view port and water refractive 
interfaces between the camera lens and the object to be 
measured. 
To determine the camera calibrations, the stereo-cameras are 
pre- or post-calibrated in shallow water, usually in a swimming 
pool, using the techniques developed by Shortis and Harvey 
(1998). The standard requirements of a multi-station self 
calibrating photogrammetric network are required, such as 
multiple convergent photographs, camera roll at each location 
and a 3D array of high contrast targets. The 3D target array, 
usually in the form of a light, easily manoeuvrable calibration 
fixture, has the size determined by the field of view of the 
cameras and the likely working distance for the measurements. 
It is impractical to manoeuvre towed body systems in the same 
way as a hand-held camera, so instead the calibration fixture is 
tilted and rotated in the field of view of the camera (see figure 3) 
to replicate the convergent multi-station network (Harvey and 
Shortis, 1996). 
The positions of the targets in the images are measured semi- 
automatically based on the centroid location of each target in 
each image. It is immaterial if the frame distorts or is dis 
assembled between calibrations, although the frame must retain 
its structural integrity during a calibration sequence. The 
results of the photogrammetric network computation for the 
self-calibration include the locations and orientations of the 
synchronised pairs of images, the calibrations of the cameras 
and revised coordinates of the target positions on the frame. 
The overall dimensional scale of the photogrammetric network 
of images and targets is determined by distance constraints 
between targets on the rigid arms of the frame. 
RMS image residuals range from 1/20 to 1/30 of a pixel for an 
in-air self-calibration network based on centroid measurements, 
dependent primarily on the target image quality and integrity of 
the calibration model (Shortis et al., 1995). The result for the 
shallow water calibrations of the towed body stereo-camera 
system is typically a RMS of no better than 1/15 pixels. The 
result is degraded compared to the equivalent result in air due to 
the impact of assumptions in the calibration model, non 
uniformities of the refractive interfaces and the dispersion of the 
water medium (Newton, 1989). The latter leads to a reduction 
in contrast, as compared to in-air images, that reduces the 
precision of the centroids. 
Figure 3. Typical images for shallow water calibration (left) 
and length validation using a known length (right). The LED is 
used for synchronisation checks.