AN EVALUATION OF THREE DIFFERENT IMAGE CAPTURE METHODS FOR MEASUREMENT AND ANALYSIS
OF DEFORMATION WITHIN A GEOTECHNICAL CENTRIFUGE
J. Chen, S. Robson, M.A.R. Cooper, & R.N.Taylor*
Engineering Surveying Research Centre
*Geotechnical Engineering Research Centre
Department of Civil Engineering, City University,
London. EC1V OHB, UK
Email:J.Chen Gcity.ac.uk
ISPRS Commission V, Working Group V/III
KEY WORDS: Acquisition, Analysis, On line, Video, JPEG
ABSTRACT
In a geotechnical centrifuge experiment, long image sequences are often captured to obtain information about how soil responds to
load. Information obtainable from the on-line dynamic deformation analysis of such sequential images can be important in order to
visualise and monitor progress and ultimately to control the course of the experiment. In this paper, the geometric qualities of three
methods which can be used for the analysis and storage of long image sequences are evaluated: (1) use of an S-VHS video recorder;
(2) use of J-PEG image compression; and (3) the on-line measurement of target image co-ordinates by an analysis procedure based
on the sequential update of prior knowledge. The precision and accuracy obtained with respect to the S-VHS video recorder and
different J-PEG compression factors are discussed. A suitable algorithm for on-line dynamic target location and deformation analysis
of the sequential images produced during centrifuge tests has been derived and successfully applied in practical experiments.
1. INTRODUCTION
Settlement can be a problem during soft ground tunnelling in
urban areas where buildings can be put at risk. Settlement and
ground movement can be simulated in the laboratory using a
geotechnical centrifuge (Taylor, 1995). The high forces
generated by the centrifuge allow not only the soil model to be
scaled down but also for a corresponding reduction in the time
for deformation to take place. The soil sample is placed in a box
with a perspex face. Targets are placed in the visible face of the
soil and their locations imaged and measured. Target
movements are traced in image space to obtain deformation
information and after appropriate transformations mathematical
models can be used to analyse the geotechnical performance of
the soil. Typical experimental duration's range from a few hours
to a couple of days. An analysis of sequential images of the soil
model under loading within the centrifuge can provide useful
information about the response of the soil. If measurements of
the imaged targets can be obtained quickly enough the
information derived can then be used to optimise the
experiment. Initial research into the sequential tracking of 400
to 700 small circular targets in centrifuge experiments has been
reported (Clarke at al, 1995).
There are several important requirements for the successful
Figure 1 Discrepancy vectors from two consecutive S-VHS
images
70
capture and analysis of sequential images in a centrifuge
experiment: rapid image grabbing; large data storage capacity
and most importantly; images must be of sufficient quality for
measurement and quantitative deformation analysis. This paper
describes comparative experiments to assess the geometric
performance of a weighted centroid target location algorithm
between direct frame grabbing, S-VHS video recording, and J-
PEG compression. A methodology for the on-line tracking of
target images leading to dynamic deformation analysis without
the need to store each image is also discussed.
2. THE CAPTURE OF IMAGE SEQUENCES WITH A S-
VHS RECORDER
A S-VHS recorder has been used to archive image sequences
during centrifuge tests, a frame grabber can then be used off-
line to capture individual images of interest. The use of a S-
VHS recorder to store images on tape has the advantage of
providing large storage capacity and the ability to capture
sequential images at camera frame rate. The accuracy of target
location from taped images has been evaluated by Hoflinger and
Beyer, 1993, and by Shortis et al, 1993. Under ideal conditions,
target location accuracy can be better than one of tenth of a
pixel. However in our case, in which black targets are inserted
into grey sand and clay soils, target location errors are typically
much larger. In this application, each imaged target is small,
Figure 2 Discrepancy vectors from two consecutive directly
grabbed images
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B5. Vienna 1996
typically oc
is also lov
backgrounc
differences
grabbing ir
discrepancy
consecutive
subsequent
Figure 2 sh
consecutive
camera witl
were stable
magnitude
caused by t
geometric
recorder is
grabbing.
To rigorous
VHS syste
different ty
Both retro-
sizes were
board bein;
for the retr
an optical I
bench and
camera. A
record the s
grabbing. A
using the S
different tar
cases, a cer
been used
illustrated i
°
8
1
RMS location errors (pixels)
o
a
S
si
~
o
8
&
sad
40 60 80
Targe
Figure 3 RMS
targets using
Figures 3 ar
different ta
targets capt
small (imag
pixels in :
procedure v
RMS target
The results
used, but t
indications.
retro-reflect
a very sma
than 1/20th
good contra
order of 1/2
or contrast |
