Geography Reference
In-Depth Information
dark beads was sufficient to make segmentation very easy.
In the two-size experiment several trials were needed
in order to choose smaller transparent beads (4 mm) to
contrast with the 6 mm dark beads (Hergault et al., 2010).
Distinct segmentation algorithms were used to detect
the two types of beads. The problem of tracking particles
along a temporal sequence is not straightforward and is
known as the point-tracking problem in the literature.
Typically, points can be either small object centroids
or interest points within larger rigid or unrigid objects.
Depending on the application and on the assumptions
made regarding point displacement, more or less complex
approaches have been proposed (Hwang, 1989; Salari and
Sethi, 1990). A particular application is to determine the
velocity field of a fluid carrying small particles (Nishino
et al., 1989; Fayolle et al., 1996; Udrea et al., 2000). In
the studies reported here, Bohm et al. (2006) and Her-
gault et al. (2010) focused on individual particle motions.
Particle velocities differed from those of the surrounding
fluid since the particles were coarse and had a higher
density than water. Although the use of spherical particles
of uniform size made detection easier, the calculation of
the trajectories was more difficult since it was not pos-
sible to distinguish particles based on their shape. The
high frame rate of the camera resulted in a displace-
ment of a particle between two images that was always
smaller than the particle diameter. This was essential to
achieving a good accuracy with the tracking algorithm as
well as to reconstructing trajectories at high resolution
(Figure 13.6b).
(a)
(b)
Figure 13.6 (a) Image of a unimodal experiment with 6 mm
beads, (b) image from an experiment with a two-size mixture,
bead-trajectories corresponding to the previous 30 images are
superimposed.
turbulent supercritical flow down a steep channel (slopes
were between 7.5-15%). Details can be found in Bohm
et al. (2004, 2006). The experiments were performed with
one-size and two-size sediment mixtures (Figures 13.6a
and b respectively). The channel was supplied contin-
uously with beads and measurements were made when
transport was at equilibrium over the mobile bed (Bohm
et al., 2004). Image processing was used to determine
velocities and trajectories of each bead from a temporal
sequence of approximately 8,000 images (Bohm et al.,
2006). Numerous measurements permitted a thorough
statistical description of unimodal sediment transport
based on stochastic Markov-type processes (Ancey et al.,
2006, 2008).
As in the previous studies, image processing consisted
of first detecting and localising all beads and then recon-
structing their trajectories by tracking them through the
sequence of images. Here again, the image processing soft-
ware WIMA in combination with several custom designed
algorithms was used. In this study all the particles present
in the measurement window were tracked (as opposed
to only coloured particles in the previous study) making
segmentation a crucial step. The quality of segmenta-
tion is largely dependent on flume setup, lighting, and
material, and requires blending of know-how on image
acquisition and knowledge of segmentation algorithms.
As a general rule, particle tracking works best when
applied to high quality well-contrasted images achieved
through the right combination of lighting and material.
Experiments should be setup with PTV needs in mind;
bad quality images result in numerous false detections
and can hamper the possibility to perform tracking at
all. In the uniform size experiment (Bohm et al., 2006),
using a two-dimensional channel slightly larger than the
13.3 Channel morphology and flow
dynamics
Timelapse imagery of dynamically evolving surfaces are
an efficient means for measuring a broad-range of vari-
ables characterising channel dynamics and bed evolution
at high temporal and spatial resolutions. Timelapse
images can be readily compiled into movies showing
the continuous evolution of a system. We present some
general considerations in acquiring timelapse imagery
and present examples on how they have been used in
experimental studies of rivers and deltas.
When acquiring timelapse images it is recommended
to use cameras that can be computer operated and allow
automatic download. Computer operation is important
because it helps prevent movement of the camera once
it is fixed in place; this is especially important if the
camera is mounted in a hard-to-reach location. Having
Search WWH ::




Custom Search