Experimental Equipment and System Construction
The major experimental equipment includes the shaking table
and the videogrammetric hardware system. A large-scale 3D
shaking table with a platform size of 4 m × 4 m was employed
for the tests. The working frequency ranged from 0.1 to 50 Hz.
The maximum driving acceleration was 1.2, 0.8, and 0.7g in
the X, Y, and Z directions, respectively.
The videogrammetric hardware system consists of the
high-speed cameras, a computer, an electronic total station,
artificial targets, etc. In this study, two synchronized
CMOS
cameras (DALSA Falcon 4M60) were placed in front of the
landslide dam model. The image resolution was 2352 (H)
× 1728 (V) pixels, and the frame rate was set at 60 frames
per second (fps). Cameras were mounted on tripods to fix
their exterior orientation parameters. The field of view of
two cameras equipped with Nikkor 20 mm f/2.8D fixed focal
length wide-angle lenses satisfies the observation’s needs. The
natural light met the requirements for the subsequent image
processing, and the lamps and the retroreflective targets are
unsuitable in this experiment as the thin tempered glass of
the test box would lead to saturation at some key positions in
the case of strong light.
Circular artificial targets representing control points and
tracking points were employed to facilitate the videogrammet-
ric measurement. As there were insufficient fixed structures
surrounding the shaking table within the field of view, steel
beams in different depths were positioned above the dam
model and steel blocks were placed in front of the model in
order to construct the control network. Control targets were
pasted on these structures and their spatial coordinates were
surveyed in advance by a SOKKIA NET05AX total station.
Reference points were selected on the test box to calculate the
relative displacement between the dam and the shaking table.
Tracking targets were regularly inserted into the dam model at
several key positions. As shown in Figure 4a and 4b, the cen-
ters of the targets were obtained using ellipse detection and
fitting, and their correspondences in the stereo pair were de-
termined by point set registration. In these figures, the circles
represent the image coordinates of targets in the left image,
and the crosses represent ones in the right image. The control
targets and tracking targets are plotted in blue and red color,
respectively. In this experiment, the parameters in the point
set registration algorithm followed the recommended settings
in the original paper (Ma
et al.
, 2016). Figure 4c shows the
layout of the targets in the first frame, which include 25 con-
trol points (denoted by C), 2 reference points (denoted by R),
and 15 tracking points (denoted by T) after target recognition
and matching. Figure 4d illustrates the exemplary template
image sequences of several tracking targets.
Results and Discussion
The performance evaluation and analysis of the proposed
subpixel phase correlation method and videogrammetric sys-
tem are reported in this section. First, the reliability of the im-
proved subpixel phase correlation method is assessed by two
simulated tests and a static test with known ground truths, in
Figure 4. The performance of point set registration, the layout of the targets and the exemplary template image sequences:
(a) The image coordinates of targets in the left image (denoted by circles) and in the right image (denoted by crosses) before
point set registration, (b) The image coordinates of targets after registering the left point set to the right point set, (c) The
layout of the targets at the first epoch, and (d) The exemplary template image sequences of several tracking targets.
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
September 2018
585
