International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B5. Istanbul 2004
was used, no quantitative evaluation about the displacements at
any other location of the terrain underneath could be derived. A
qualitative information was obtained by inserting coloured
horizontal sandy layers and taking some pictures during trials.
B
Figure 1: The loading device and the experimental setup
2.3 Design of the photogrammetric survey
In order to acquire more accurate information about the
displacement field under the foundation, a digital camera has
been placed in quasi-nadir position with respect to one of the
glass walls, recording images at a rate allowing to describe the
phenomenon until the collapse of the sand.
Because of the test setup, the displacement field can be
considered as plane (and symmetrical with respect to the
vertical plane normal to the glass and passing through the
foundation centre). This allows to apply a image rectification to
convert image displacement into object displacements, with the
additional advantage of a simpler setup (a single camera, no
synchronization required).
To compute the 8 parameters of the transformation for every
image of the sequence and to establish a permanent reference
system to make comparisons, a polyester frame has been placed
on the outer side of the glass, covering the area interested by the
displacements (20x30cm). A set of 32 calibration targets, whose
position is known with an accuracy of about 0.01 mm has been
printed on it (see Figure 2). Computation of the parameters for
each image of the sequence would be redundant if the
arrangement camera-device is perfectly stable during the test;
unfortunately, this turned out not to be true after the first
experimental trials, which revealed small movements of the
camera, most probably due to vibrations caused by the
compressor needed to operate the oleo-dynamic actuators.
In principle, the transformation parameters computed using
points on the reference frame should not be applied to points of
the of sand layer on the opposite side of the glass wall, because
they lay on a different plane. However, the total displacement
of a point during the test is just of a few mm and only
displacements, and not absolute coordinates, must be measured
precisely. Therefore the error coming from ignoring refraction
through the glass and applying the rectification parameters to a
different plane has been neglected. By a similar reasoning, also
lens distortion has been neglected, since its variation along the
image displacement of a point is very small.
A Kodak DCS420 with a CCD array of 1524 by 1012 pixels
(square pixels with 9 um size) has been fitted with a 24 mm
lens. Because each loading test lasts 15-30 minutes, an external
electronic trigger has been connected to the camera in order to
control image acquisition at fixed steps, synchronized to
transducers and other sensors for comparison. The frequency of
acquisitions depends on the speed of displacement in sand; for
tests described below, four images per minute have been taken,
even though only a subset of these has been processed.
The camera is placed on a tripod at a distance of about 50 cm
from the glass wall, resulting in a mean pixel-size in object
space of about 0.2 mm.
As shown in the following, after an initial setup for the test, the
measurement proceeds automatically, driven by image
matching algorithms. The extraction and measurement of
calibration targets is carried out by l.s. template matching, being
a priori known the shape of each target to be measured in all
images. Tracing of sand's surface displacements consists in
finding the same homologous points in all the image sequence.
Because of their size and of movements induced by the stress
(that lead to a certain mixing, rearranging and rotating them)
neither individual grains nor cluster of them can be traced.
Several small white spheres have therefore been positioned as
tracing points in the sand near the glass wall, at the nodes of a
grid (see figure 2). Weight and diameter of spheres has been
selected to let unmodified the mechanical behaviour of sand,
while their colour contrasted enough to guarantee identification
against the background. :
Although preparation of the specimen is not a simple and fast
matter (some hours for each loading test are needed), this
solution yields a high resolution description of the displacement
field. This advantage is very important for the geotechnical
analysis, where investigation of initial loading steps is the most
interesting issue. Because at this stage displacement are very
small, the accuracy required is in the order of 0.1 mm at object
scale.
Figure 2: The reference frame and the area of the specimen
interested by displacements (about 30x20 cm),
before and after loading (test 1).
2.4 Processing of a test image sequence
Once a test has been completed, images are downloaded from
the camera memory to a PC. The quality of the colour images 18
generally not very good with this camera, but no pre-processing
Internation
AB
stages are
conversion
a well co
reflective,
a way as t
sand.
To control
to setup th
out some 1
labelling
measurenk
calibration
fiducials 1
conformal
reference «
following
can be use
perfectly s
and appro»
Tracing the
following |
LI BD
PN rM fF ın
Cs
Image me
(Gruen, 1%
compensat
corresponc
that the sp
allowing
sometimes
sand, so al
No radion
algorithm,
equalized |
The meası
template n
tracing pc
around the
this windo
image. Th
small, as i
later the m
than the
Modelling
would all
basis of it
next secti
exploits t|
displacem
positions c
This metl
sequence,
stages of
compromi
cause of f.
into the s
glass or I:
simple pre
