vould be
ed to this
:d system
of points
1e surface
manual
reeze the
ammetric
bility of
«digital
te natural
tural form
n the field
is. judged
used for
94) were
ly faceted
ny power
et (Figure
have been
sed retro-
reflective balls floating on partially filled inner-tubes to record
the oscillating water surface within a canal lock. Various other
designs involving freely floating particles and fishing nets were
considered before deciding upon a series of black polystyrene
balls constrained by fine lines. The spherical nature of the
sphere is of course important, providing the same circular
target when viewed from any point in space. The target colour
was significant also, the pro-glacial stream would be heavily
laden with finely ground rock flour and this creates a milky
white appearance to the water in this area of the Swiss Alps.
Once the images had been scanned it was hoped that the black
polystyrene ball could be identified readily using a simple grey
level threshold. Prior knowledge concerning the likely scale of
the images and planned scanning resolution yielded an ideal
target design diameter of 30mm.
Target tests were conducted in a laboratory flume prior to
carrying out the fieldwork. The water flowed at a velocity of
Ims! and these tests identified a critical problem and a simple
but appropriate solution. When the targets were held in a fixed
position within the flume, the high flow velocity created a
noticeable wake and wave of water flowed over the ball. The
wake both obscured the ball caused the ball to sink into the
water, thereby creating a more significant and complex
systematic error than the envisaged simple and minimal
vertical offset due to ball weight. By allowing the balls to flow
freely with the water during the instant that the photographs
were acquired, it was found that the wake disappeared and the
balls clearly floated freely upon the water surface.
Photographic tests and consequent scanning and centroid
measurement using image processing methods suggested that
the targeting system would indeed be successful.
The UK fishing community provided a convenient and cost
effective supply of all raw materials used to construct the
targets. Black "Poly-balls" of various diameters (12mm, 18mm,
25mm, 37mm) are used in the UK by carp and pike anglers
and one hundred 25mm and 37mm diameter Poly-balls were
purchased. 100m of fishing line (breaking strain of 121bs) and
split shot were used to fix the positions of the target poly-balls
at pre-defined locations on the line. Total cost of all materials
was less than $US 75. The final design used on site involved
the use of six fishing lines each with eight targets placed at
0.5m intervals. The lines were attached to a 2.0m length of
plastic piping, each line spaced 0.6m apart. The whole net’ of
points could be moved to the required locations within the
confluence by two researchers clad in dry-suits (Figure 1).
Although the targeting of points to mark the water surface
provided the main technical problem for the design of the
system, it was necessary to provide some form of target to
provide photo-control for the photogrammetry. Painted boards
of dimensions 0.8 x 0.8m and smaller targets 0.06 x 0.04m
were used to provide such photo-control points, and were
installed on the sides of the stream bank (Figure 1). One minor
problem experienced was the loss of one of these boards due to
the rise in water level during the afternoon melt period. The
3D coordinates of the photo-control points were established
using a conventional theodolite intersection method using a
Geodimeter 410 digital tacheometer. Data were recorded
electronically, subsequently downloaded and processed using a
least squares '3D variation of coordinates program. This
provided the best estimates for the 3D coordinates for these
points and associated estimates of precision.
2.2 Imaging
Two modified Hasselblad ELX small format cameras were
used to record images suitable for photogrammetric processing.
These were both equipped with a 25 cross glass reseau plate
set in the focal plane of the cameras. The location of each cross
had been measured to a precision of +1 micron and could be
regarded as sufficiently stable to define a rigid photo-
coordinate system. The lens could be pinned at any desired
focal length using three grub-screws and could also be
regarded as stable. These two features enable the camera to be
classified as a 'semi-metric' camera and appropriate calibration
parameters to model inner orientation had been determined
using a test field constructed at Loughborough University
(Chandler & Padfield, 1996). The cameras featured a 220 film
back which could accommodate both 120 and 220 roll film,
although 220 film was finally used. Ilford FP4 emulsion was
utilised and check processing carried on site, it was felt
essential to verify the adequacy of the selected exposures so
that results could be guaranteed.
The two cameras were electrically synchronised so that
coincident exposures could be obtained. The cameras were
positioned on conventional camera tripods, on top of a small
scaffold platform situated on the stream-bank. It had been
hoped to position the platform and hence cameras above and
across the stream flow, but high flow rates prevented this.
Upon return to the UK the remaining films were processed and
the original negatives scanned at a resolution of 20 microns
using the Helava DSW100 Digital Scanning Workstation at
City University, London. Each image produced a file which
was approximately 8.2Mb in size.
2.3 Automated measurement
The image coordinates for the surface targets were measured
using automated methods provided by the Visilog image
processing package. Visilog provides a modular approach to
image processing and analysis, and is supported on a variety of
platforms including Microsoft Windows and UNIX. The
package includes a comprehensive suite of image processing
libraries including: convolution and spatial filters; edge
detection, linking and approximation; frequency domain
processing and significantly global analysis of objects
following application of a labelling technique (Boyle &
Thomas, 1988). This latter feature includes a series of
measurements' which can be obtained for all detected particles
or objects, two of which include the centre of gravity (Xand
YgreyFirstOrderMoments) derived from the original grey scale
image. Visilog is equipped also with a C interpreter which
allows the user to record, develop and refine sequences of
image processing operations necessary to carry out a particular
task. The package had been used successfully to measure the
centroids of a targets prior (Chandler and Padfield, 1996) but
the original sequence of operations had to be modified slightly
for this particular measurement task. Application of the
original algorithm generated approximately 4,000 measured
particles when it was known that only 50 targets were present.
The method used to isolate the required points involved
measuring the size and shape of all particles in addition to the
centre of gravity. A simple program then isolated the particles
which were likely to have been created by the floating surface
markers based upon expected size and shape. The technique
proved to most effective, reducing the number of image
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996
