Static Test
In the static test, the 60 initial stereo frames with null
displacement before the start of the shaking table test were
employed. Two tracking points that were partially affected
by sand and two well-shaped tracking points were selected
as the test positions, and were tracked using the proposed
method and three other popular videogrammetric tracking
methods, i.e., the widely used normalized cross-correlation
(NCC) with parabola interpolation (e.g., as adopted in Chang
and Ji (2007) and Sładek
et al.
(2013)), iterative optimization-
based Lucas-Kanade (
LK
) tracking (e.g., as adopted in Ji and
Chang (2008) and Guo and Zhu (2016)) and shape-based
tracking (e.g., as adopted in Liu
et al.
(2015)). As the rela-
tive displacements were theoretically 0, the
RMS
value of the
tracking results was regarded as the tracking error. Table 3
lists the tracking errors of the four test positions for the four
different methods in the left and right image sequences. It can
be seen that the correlation-based tracking method is more
stable than the shape-based method in the corrupted case,
and the proposed method is the most reliable, achieving the
minimum tracking errors in all cases.
Results and Analysis of the Videogrammetric Measurement
The active time of the tests was around 8 s, and the result-
ing 480 stereo images were employed to reflect the dynamic
response of the landslide dam model on the large-scale
shaking table. A total of 13 tracking points and 2 reference
points, whose positions are plotted in Figure 4c (T3 and T10
are invalid due to serious occlusion by sand), were finally
processed using the proposed videogrammetric system.
Figure 6 presents the filtered deformations with respect to
reference point R1 and the accelerations of the tracking points
in the X, Y, and Z directions, respectively. It can be seen from
Figure 6a, 6b, and 6c that the dam deformation mainly lies
in the X and Z directions. The final X-direction deformations
on the positions of the tracking points in the left part move
leftwards, and vice versa. The Z-direction deformations repre-
sent the settlement, and basically meet the physical law that
the magnitude increases from the dam bottom to the crest.
As shown in Figure 6d, 6e, and 6f, the accelerations of all
the tracking points in the X direction, which is the dominant
vibration direction, share a similar tendency, while the ac-
celerations in the Y and Z directions are small and irregular,
which may be due to the influence of measurement noise,
even though it has been greatly suppressed by the use of the
Savitzky-Golay filter.
In order to assess the performance of the videogrammet-
ric measurement, 10 of 25 control points were selected as
checkpoints and were not used in the bundle adjustment. The
absolute and relative discrepancy between the videogrammet-
ric measurement and the total station were estimated using
Figure 6. The curves of the filtered relative displacement and acceleration of the tracking points: (a, b, and c) Relative
displacement in the X, Y, Z directions, and (d, e, f) Acceleration in the X, Y, Z directions.
588
September 2018
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
