International Journal of Sediment Research 29 (2014) 269-277

Pier scour monitoring system by bed-level image tracking

Wen-Yi CHANG1, Jihn-Sung LAI2, Teng-Yi YU3, Franco LIN4,
Lung-Cheng LEE5, Whey-Fone TSAI6, and Chin-Hsiung LOH7

Abstract
A scour monitoring system with a micro camera tracking the bed-level images is proposed in this study. Two
image recognition algorithms have been developed to support the bed-level image tracking approach. Through the
laboratory experiments of pier scour, this study demonstrates that the proposed system is able to accurately
monitor the scour-depth evolution in real time. In addition, five commonly-used temporal scour models are
employed to simulate scour-depth evolution and their results are compared with monitoring data. In general, the
results indicate that the proposed scour monitoring system has the potential for further applications in the field.

Key Words: Pier scour, Scour monitoring, Image recognition, Bed-level image tracking, Scour prediction

1 Introduction
The major causes of bridge failures are resulted from severe pier scour during flood periods (Melville and Coleman,
2000; Coleman and Melville, 2001). Due to lack of bridge safety warning systems, bridge failures usually caused
serious damages to people’s lives and properties (Wang et al., 2010). Thus, the technology of pier scour monitoring
during floods is very crucial to capture the in-situ scouring scenario. This will be the strong check point to determine the
timing of bridge closure.
In the past, there have been quite a few studies focused on the development of scour monitoring techniques or systems.
Forde et al. (1999) made a measurement of bridge scour by using ground penetrating radar (GPR) in a shallow
freshwater environment; the limitation in use was that GPR would not effectively penetrate the salt water with a
concentration larger than 0.05%. Yankielun and Zabilansky (1999) developed a time-domain reflectometry (TDR)
system to measure the scour depth around the piers. Due to the attenuation and dispersion of TDR signal, it might be
difficult to detect the subtle scour changes. Fukui and Otuka (2002) developed a scour monitoring method by using a
ring with magnetic sensors attached on the pier to obtain the maximum scour depth during floods. However, this
method cannot account for the refilling of sediment. Falco and Mele (2002) presented the scour monitoring results by
sonar at two bridges on the River Po. It showed that the system was able to monitor the scour level during floods;
however, the air bubbles or suspended sediment particles in the water may cause abnormal sonar reflection. Lin et al.
(2005) developed a scour monitoring system using fiber Bragg grating (FBG) sensors to real-time measure the scour
variation in the laboratory experiments. The results showed that the measured strain has a high correlation with the
scour depth; however, the fluctuation of strain measurement was quite large to limit the measurement precision. Lu et al.
(2008a) used the equipments of a sliding magnetic collar and a steel rod for the local scour measurement, and a
numbered brick column for the measurement of upstream riverbed gradation to collect bridge scour data during a flood.

1
Researcher, National Center for High-Performance Computing, National Applied Research Laboratories, Taiwan, China, E-mail:
[email protected].
2
Research fellow, Hydrotech Research Institute, National Taiwan University, Taiwan, China, E-mail: [email protected].
3
Assistant Researcher, National Center for High-Performance Computing, National Applied Research Laboratories, Taiwan,
China, E-mail: [email protected].
4
Assistant Researcher, National Center for High-Performance Computing, National Applied Research Laboratories, Taiwan,
China, E-mail: [email protected].
5
Associate Researcher, National Center for High-Performance Computing, National Applied Research Laboratories, Taiwan,
China, E-mail: [email protected].
6
Senior Researcher, National Center for High-Performance Computing, National Applied Research Laboratories, Taiwan, China,
E-mail: [email protected].
7
Prof., Department of Civil Engineering, National Taiwan University, Taiwan, China, E-mail: [email protected]
Note: The original manuscript of this paper was received in Aug. 2012. The revised version was received in Oct. 2013. Discussion
open until June 2015.
International Journal of Sediment Research, Vol. 29, No. 2, 2014, pp. 269–277 - 269 -
Due to the potential limitation of gravity-type measurement, their system couldn’t consider the refilling process of
sediment. Lin et al. (2010) applied the micro-electro-mechanical system (MEMS) pressure sensors installed on the
cylinder to real-time monitor the scour depth under an unsteady hydrograph. It was found that only the pressure sensor
near the bed level would effectively activate, hence, the measurement of scour-depth evolution may be discrete
depending on the sensor deployment. From the above literature review, it is found that most of these available
techniques have limited applications. For instance, although sonar and radar are easy to install, but the results are
difficult to interpret, especially when the flow contains high concentration sediments, bubbles or debris during a flood.
On the other hand, the gravity-type monitoring systems seem to give a stable scour measurement. However, the
limitation practically makes it hard to measure the refilling process of sediment. In addition, the accuracy of scour
measurement is still not satisfactory.
In the present study, we developed a high-accuracy scour monitoring system utilizing a micro camera monitoring and
tracking bed-level images. Through a series of scour image processing, the system can real-time recognize the bed-level
position and obtain the scour-depth evolution. Hence, the automation of measuring procedure is implemented. To our
knowledge, this is the first time that the bed-level image tracking techniques are used to monitoring pier scourred
automatically in real-time. In addition, since the scour images and monitoring scour depth can be obtained
simultaneously, the intuitive information of scour images would be very helpful for operators to verify the monitoring
data. The proposed scour monitoring system is applied in the laboratory experiments to validate its capability and
applicability. Then, five commonly-used scour models are employed for comparison with the scour monitoring data to
investigate these models’ applicability.

2 Scour monitoring system

2.1 Framework
The framework of the present scour monitoring system is shown in Fig. 1. The equipments are set up in a transparent
hollow cylinder as a pier, which includes a rail with a moving holder on it. A micro camera is mounted on the movable
holder which can be driven by the motor to move on the rail. The camera is employed to monitor and take the images of
scouring bed level (i.e. scour line) around the pier. Then, the images are sent to the remote computer processor for
scour-line image recognition. The process of image recognition will be elaborated in Section 2.2. Once the scour line on
the image is recognized and found to exceed the prescribed range, the system will send a signal through wireless
communication to the motor to move the holder upward or downward, focusing on the scour position. Therefore, the
scour line is always located right in the monitoring screen of the camera. A prototype of the present scour monitoring
system was developed and applied to the laboratory experiments for pier scour monitoring. A graphical user interface
(GUI) is developed to assist users in the system operation. Both the real-time monitoring images and the recognized
scour-depth evolution can be displayed on the GUI system screen.

2.2 Image recognition algorithms

According to the observation of the sequential scour images, two important characteristics of scouring process can be
utilized to implement the image recognition of scour depth. The first characteristic is the sharp brightness variation
around the scour bed level, and the corresponding image recognition technique is the Brightness Intensity Segmentation
(BIS) method to obtain the scour line. The second characteristic of the scouring process is the active motion of sediment
particles above the scour bed level, and the Particle Motion Detection (PMD) method can be employed to recognize the
scour line.
For the BIS method, the image processing techniques consisting of noise reduction, morphological calculation, image
threshold calculation, and edge detection (Gonzalez and Woods, 2002; Lu et al., 2008b) are adopted in the present study.
The procedure of scour-line recognition of scour images is presented in Fig. 2. To begin with, as shown in Figs.
2(a)~2(b), the original scour image is transformed to the gray-level image and reduced noise by the Gaussian filter with
a 7 × 7 (pixels) filtering element. According to the characteristics of sediment entrainment in the scour hole around the
pier nose, the scour position represents the location at which a boundary line separates the scour bottom along the pier
from the water. In Fig. 2(b), one can see that the area above the scour position is much brighter than those in the area
below it. In other words, the darker part represents the immovable layer of sediment beneath the bed surface. By
utilizing this characteristic, the scour image is further processed by the horizontal Sobel Filtering (Gonzalez and Woods,
2002) to enhance the horizontal intensity, as shown in Fig. 2(c). Then, the Otsu’s binarization algorithm (Otsu, 1979) is
adopted to identify the boundary line from the image, as shown in Fig. 2(d). It is found that some small black patches
existing above the scour position. Therefore, as shown in Fig. 2(e), the morphological calculation (Haralick et al., 1987)
is utilized with a 7 × 7 (pixels) structuring element to eliminate these unwanted noise. Furthermore, as shown in Fig.
2(f), Hough transformation (Duda and Hart, 1972) is utilized to extract the straight line of scour position, and then the
scour line is plotted on the original image for comparison. In Fig. 2(f), one can see that the recognized scour line (green
line) agree well with the observed one to demonstrate good tracking ability of the scour position by the BIS method.

- 270 - International Journal of Sediment Research, Vol. 29, No. 2, 2014, pp. 269–277
 Fig. 1 Framework of the scour monitoring system

Fig. 2 Processing images for the BIS method: (a) original image, (b) gray-level image, (c) horizontal Sobel filtering image,
(d) Otsu’s binarization image, (e) morphological calculation image, and (f) recognized scour-line image

For the PMD method, the procedure of scour-line recognition of scour images is presented in Fig. 3. In the beginning,
as shown in Figs. 3(a)~3(b), two sequential scour images are captured with 1-second time difference and transformed to
the gray-level images. According to the characteristic of active particle motion above the scour bottom driven by the
horseshoe vortex flow motion in front of the pier, the particle motion region can be treated as the foreground while the
immovable bed layer is treated as the background. To extract the foreground from the scour images, the gray-level
difference is calculated and binarized by a threshold value. As shown in Fig. 3(c), the massive white patches
representing the foreground indicate the active motion of sediment particles. In order to enhance the foreground
intensity, ten to twenty foreground images are accumulated by using the scoreboard calculation. Figure 3(d) shows the
enhanced foreground image by accumulating twenty consecutive foreground images. One can see that the interface of
the white foreground and black background clearly separates the scour bottom along the pier from the water. Then, as
shown in Fig. 3(e), the morphological calculation is utilized to eliminate the unwanted noise. Furthermore, as shown in
Fig. 3(f), Hough transformation is adopted to extract the straight line of scour position, and then the scour line is plotted
on the original image for comparison. In Fig. 3(f), one can see that the recognized scour line (red line) agrees well with
the observed one to demonstrate good tracking ability of the scour position by the PMD method.
International Journal of Sediment Research, Vol. 29, No. 2, 2014, pp. 269–277 - 271 -
 Once the scour line is recognized by the BIS method or the PMD method, the actual scour depth can be calculated by
the following algorithms. As shown in Fig. 4, initially the initial scour line is placed on the center line of the image by
manipulating the camera holder. Herein, Pc represents the number of pixels counted from the up-boundary of the image
for the center line, and P (t = 0) = Pc . As scour occurs, the scour line moves to the position of P (t ) . In order to avoid
the position of scour line going beyond the extent of the monitoring screen, two reference lines are employed. For each
monitoring image, when the scour line goes below the Low-bound line, the holder will automatically move the camera
down by a distance of (PL-Pc), and vice versa. Therefore, the scour depth can be calculated by the following relation:
L (1)
d s = [(P (t ) − Pc ) + N L (PL − Pc ) + NU (PU − Pc )] ⋅ H
PH
where ds is the scour depth; P (t ) is the position of scour line in terms of pixel at time t; N L and NU are the
numbers that the holder moves the camera down or up, respectively; LH is the height of the monitoring image; and
PH is the total pixel number of the image height. In the present study, the dimension of the monitoring image is
320*240 pixels, and PH = 240 pixels. Furthermore, according to the reference ruler attached on the monitoring screen,
LH is calibrated to be 3.8 cm. By using Eq. (1), the system can obtain a real-time output of the scour-depth variation
data based on the recognized scour-line images. In this study, the time-discretisation of the scour depth tracking in the
experiments is set as 1 second.

Fig. 3 Processing images for the PMD method: (a) original image, (b) gray-level image, (c) motion detection image, (d)
scoreboard calculation image, (e) morphological calculation image, and (f) recognized scour-line image

In the present study, only the BIS method is incorporated into the present scour monitoring system to test directly in
the laboratory experiments. However, one of the experimental results is employed to investigate the capability of the
PMD method under the live-bed scour condition. In fact, the PMD method is found quite suitable for the scour-line
recognition in the high turbid flow.

Fig. 4 Illustration of scour-line location on the recognized image

- 272 - International Journal of Sediment Research, Vol. 29, No. 2, 2014, pp. 269–277
3 Experiments and results

3.1 Experimental setup

The experiments were conducted in a flume with 36 m in length, 1 m in width, and 1.1 m in depth. The flume had
glass sidewalls, and is located at the Hydrotech Research Institute of National Taiwan University, Taipei, Taiwan. In the
test reach of the flume, a false floor is used with a 2.8 m long, 1 m wide and 0.3 m deep recess in which the sediment is
placed. The recess is filled by nearly uniform sediment having a median size of 0.86 mm with the geometric standard
deviation of sediment size distribution σ g = 1.11 . A hollow cylindrical pier made of Plexiglas with a diameter of 0.11 m
is located at the middle of the recess. The present scour monitoring system is installed in this hollow cylindrical pier to
measure the scour position outside of the pier. The prescribed inflow discharge and its corresponding depth for the
experiments are controlled by adjusting the inlet valve and the tailgate. In the present study, three pier scour
experiments have been carried out with different flow intensity conditions. Two of them are under the clear-water scour
condition for the scour model calibration and verification. The remained one is under the live-bed scour condition and
employed to investigate the capability of the PMD method for the scour-line recognition in the high turbid flow. Table 1
lists the experimental conditions for three cases.

Table 1 Experimental conditions

Run do(mm) σg H(m) V(m s-1) Q(m3 s-1) V/Vc td(hr) ds/D
Case 1 0.86 1.11 0.250 0.300 0.075 0.77 7 1.10
Case 2 0.86 1.11 0.178 0.352 0.063 0.95 7 1.27
Case 3* 0.86 1.11 0.220 0.404 0.089 1.06 7 1.25
*
No sediment supply in the upstream

3.2 Experimental results and discussion

In the experiments, when the prescribed flow condition was reached, the scour monitoring system was activated. Then,
a thin Plexiglas sheet covering sediment around the pier was removed carefully to begin the experiment. We have
learned that the scour rate at initial scouring stage is highest during the experiment (Chang et al., 2004). The scour depth
in the first 10% of the time to equilibrium would contribute about 50~80% of the equilibrium scour depth (Melville &
Chiew, 1999). Therefore, the present scour monitoring system needs to capture the rapid variation of scour depth at
initial stage for practice applications. In the present study, the time interval of scour image recognition was set to 1
second throughout the experiments. Taking Case 2 as an example, Fig. 5 shows the real-time monitoring scour images
and scour-depth recognition results at different time. In Fig. 5(a), before the removal of the covering sheet from the bed,
the initial scour line was place on the center line of the image by manipulating the camera holder. After removing the
covering sheet, the scour line was captured definitely in the scour monitoring system as shown in Fig. 5(b). Since the
scouring process was rather rapid at the initial stage, the recognized scour line moved down quickly to be below the
low-bound line (as shown in Fig. 5(c)). At this moment, the system sent commands to the motor and moved the camera
downward to track the bed level. In Fig. 5(d), one can see that the camera had been moved to the position having the
scour line on the center of the image. Through the experiment, the scour line was well captured by the proposed image
recognition algorithm (BIS method), and the scour-depth evolution was successfully measured by the present system.
Fig. 6 presents the monitoring data of scour-depth evolution compared with the visual data for three cases. Herein, the
visual data are obtained by examining the video records of each experiment. In Fig. 6, Cases 1 and 2 are corresponding
to the low flow intensity (V/Vc=0.77) and the high flow intensity (V/Vc=0.95) under the clear-water scour condition,
respectively. Case 3 is supposed to correspond to the turbid flow (V/Vc=1.06) under the live-bed scour condition. One
can see that the monitoring data coincide with the visual data very well in all three cases to show good capability and
applicability of the present scour monitoring system incorporated with the BIS bed-level image tracking method.
However, during a flood in the field, it is usually considered that the high amounts of suspended load may make a visual
bed boundary hard to determine in the turbid flow. In that case, the PMD method may be an alternative for the
scour-line recognition in the high turbid flow; hence Case 3 is employed for the validation of the PMD method. Figure 7
shows the recognized scour-depth evolution by the PMD method compared with other methods. Good matching of the
monitoring results shows good promising for the PMD method. Therefore, for the bed-level image tracking in the fields,
development of an integrated scour monitoring system consisted of the BIS method and PMD method will be the next
step of the present study.

3.3 Scour simulations

In addition to scour monitoring, scour simulation is usually the alternative way to provide the detailed pier scour
information during floods. For engineering design purposes, the regression models have been developed for estimating
the equilibrium scour depth at bridge piers (Shen et al., 1969; Raudkivi and Ettema, 1983; Chiew and Melville, 1987;
Melville and Coleman, 2000). In recent years, the modern data-driven modeling tools based on the artificial intelligence

International Journal of Sediment Research, Vol. 29, No. 2, 2014, pp. 269–277 - 273 -
 Fig. 5 Real-time monitoring scour images and scour-depth recognition results at different time for
Case 2: (a) before the removal of the covering sheet, (b) after the removal of the covering
sheet, (c) at time=10 min. 0 sec. and (d) at time=10 min. 20 sec

Fig. 6 Monitoring data of scour-depth evolution compared Fig. 7 Comparison of the recognized scour-depth evolution
with the visual data by the different algorithms

(AI) techniques have made good progress. These AI-based techniques, such as GEP-based models and ANN-based
models, have been found excellent applications for bridge scour prediction in the fields (Firat, 2009; Amini et al., 2011;
Guven et al., 2012; Khan et al., 2012a; Khan et al., 2012b; Azamathulla, 2012). In addition, when the flow unsteadiness
is pronounced, the temporal effect on scour depth should be considered. In the literatures, several temporal scour
models (Kothyari et al., 1992; Melville and Chiew, 1999; Oliveto and Hager, 2002; Mia and Nago, 2003; Chang et al.,
2004) have been developed under clear-water scour conditions. In this study, these temporal scour models are employed
for simulation and comparison with the scour monitoring data for studying their capability and applicability. The details
of experimental conditions are listed in Table 1, in which Cases 1 and 2 are employed for the model calibration and
verification, respectively.
For the model calibration, the parameters calibrated in each scour model are described below. For Kothyari et al.’s
model (1992), the repose angle of sediment is assumed to be 33 o according to the sediment size d o used. For
Melville and Chiew’ model (1999), all model parameters for simulation can be determined by the given relations. For
Oliveto and Hager’s model (2002; 2005), the threshold Froude number Ft is calculated to be 0.73, and the shape factor
N is equal to 1 because the circular pier is considered. In addition, as mentioned in the original study (Oliveto and
- 274 - International Journal of Sediment Research, Vol. 29, No. 2, 2014, pp. 269–277
Hager, 2005), the parameter CFt may change from 0.068 to 0.048 depending on the threshold Froude number Ft .
Therefore, both values of CFt are adopted for simulation and comparison in Case 1, and the calibrated one is used in
Case 2 for verification. For Mia and Nago’s model (2003), like Kothyari et al.’s model, the repose angle of sediment
φ = 33o is used. For Chang et al.’s model (2004), the parameter Kσ is taken 1 because the uniform sediment is used.
The parameters K d and K h , representing the sediment size and water depth effects on the equilibrium scour depth,
are determined by the relation proposed by Chiew and Melville (1987).
For the model calibration, Fig. 8(a) shows the simulation of scour-depth evolution compared with the measured data
for each model. In the figure, one can see that the simulated result by Kothyari et al.’s model is slightly less than the
measured data. It seems that simulations of Kothyari et al.’s model tend to underestimate the scour depth variation in the
literature (Mia and Nago, 2003). For Oliveto and Hager’s model, both simulations of scour-depth evolution with
CFt = 0.068 and CFt = 0.048 are plotted in Fig. 8(a) for comparison. One can see that, with CFt = 0.068 , the simulated
scour-depth evolution is much higher than the measured data. However, if CFt = 0.048 is adopted, the simulated
scour-depth evolution gets closer to the measured data. Therefore, CFt = 0.048 is taken as the calibration result, which
is also used in the literatures (Oliveto and Hager, 2005; Lopez and Dias, 2006). For Melville and Chiew’s model, one
can see that the simulated scour-depth evolution is much higher than the measured data. In fact, their method is mainly
based on the analysis of a large amount of experimental or field data (Melville, 1997). For the safety design purposes,
their regressive scour formula may approach the enveloped curve of the employed data. Hence, the simulated results
carry out higher values than those obtained from the experiment. For Mia and Nago’s model, also considering the
concept of the primary vortex in front of the pier, the simulated scour-depth evolution is found to be very close to the
measured data. Table 2 lists the scour-depth deviations between the simulated and measured data at each hour of the
experiment for each model. In general, it is found that Mia and Nago’s model produces the best result of the scour-depth
evolution in Case 1, in which the average deviation of scour depth is 0.026 D.

Fig. 8 Simulations of scour-depth evolution compared with the measured data: (a) model calibration, (b) model verification

For the model verification, all the parameters used in each scour model for Case 2 are determined by either the
experimental conditions or the calibration results. Figure 8(b) shows the simulations of scour-depth evolution compared
with the measured data for Case 2, in which the calibrated C Ft = 0.048 is used in Oliveto and Hager’s model. In the
figure, except Melville and Chiew’s model, all simulated scour-depth evolutions coincide with the measured data quite
well. Again, the scour-depth deviations between the simulated and measured data at each hour are listed in Table 2 for
comparison. It is found that except Melville and Chiew’s model, all the average deviations of scour depth are less than
0.07 D, and Mia and Nago’s model also produces the best simulation result of the scour-depth evolution in Case 2.

4 Conclusions
A pier scour monitoring system by bed-level image tracking has been developed and validated through laboratory
experiments. From the results of the present study, the following conclusions are drawn:
(1) Considering the characteristics of the scouring process in front of the pier, the brightness intensity segmentation
(BIS) method and the particle motion detection (PMD) method are developed for the bed-level image recognition,
which help to implement the automation of the monitoring procedure.
(2) The laboratory experiments of pier scour were conducted to validate the proposed scour monitoring system. The
results show that monitored scour-depth evolutions coincide with the visual data very well. This result demonstrates that
the high accuracy and strong capability of the proposed system. In addition, the intuitive information of scour images
would also help the system operator to directly verify the monitoring data.
International Journal of Sediment Research, Vol. 29, No. 2, 2014, pp. 269–277 - 275 -
 Table 2 Scour-depth deviations between the simulated and measured data
Model Case 1st hr 2nd hr 3rd hr 4th hr 5th hr 6th hr 7th hr Ave.
1 0.148 0.129 0.153 0.160 0.182 0.187 0.202 0.166
Kothyari et al. (1992)
2 0.058 0.059 0.060 0.062 0.060 0.064 0.083 0.064
1 0.292 0.339 0.331 0.334 0.316 0.316 0.302 0.319
Melville and Chiew (1999)
2 0.556 0.571 0.572 0.567 0.565 0.556 0.530 0.560
1 0.131 0.113 0.065 0.040 0.001 0.017 0.046 0.059
Oliveto and Hager (2002)-CFt=0.048
2 0.122 0.078 0.045 0.017 0.003 0.026 0.062 0.050
Oliveto and Hager (2002)-CFt=0.068 1 0.484 0.497 0.467 0.455 0.425 0.416 0.394 0.448
1 0.018 0.046 0.023 0.014 0.013 0.022 0.043 0.026
Mia and Nago (2003)
2 0.008 0.011 0.003 0.008 0.014 0.028 0.057 0.018
1 0.138 0.129 0.079 0.050 0.006 0.013 0.044 0.066
Chang et al. (2004)
2 0.087 0.081 0.059 0.034 0.015 0.004 0.041 0.046

(3) For scour simulations, five commonly-used scour models are employed for comparison with the measured data
under the clear-water scour condition. The model calibration and verification show that Mia and Nago’s model gives the
best simulation results of the scour-depth evolution in this study.
(4) In general, the experimental results indicate that the real-time monitoring system using the bed-level image
tracking methods has the potential for further applications in the field. However, turbid flow may scratch the surface of
the monitoring tube to reduce its transparency after floods. Therefore, regular maintenance is required to keep the
monitoring surface clear for long-term operation.

Acknowledgements
Financial support from the National Science Council, Taiwan, under grants NSC 98-2625-M-002-018-MY3 and NSC
101-2625-M-492-006 is highly appreciated. We are also grateful to the National Center for High-Performance
Computing and the Hydrotech Research Institute of National Taiwan University for their computing and experimental
facilities.

References
Amini S. A., Mohammad T. A., Aziz A. A., Ghazali A. H., and Huat B. B. K. 2011, A local scour prediction method for pile caps in
complex piers. Proceedings of ICE-Water Management, 164, No. 2, pp. 73–80.
Azamathulla H. Md. 2012, Gene-expression programming to predict scour at a bridge abutment. IWA Journal of Hydroinformatics,
Vol. 14, No. 2, pp. 324–331.
Chang W. Y., Lai J. S., and Yen C. L. 2004, Evolution of scour depth at circular bridge piers. Journal of Hydraulic Engineering, Vol.
130, No. 9, pp. 905–913.
Chiew Y. M. and Melville B. W. 1987, Local scour around bridge piers. Journal of Hydraulic Research, Vol. 25, No. 1, pp. 15–26.
Coleman S. E. and Melville B. W. 2001, Case study: New Zealand bridge scour experiences. Journal of Hydraulic Engineering, Vol.
127, No. 7, pp. 535–546.
Duda R. O. and Hart P. E. 1972, Use of the Hough Transformation to Detect Lines and Curves in Pictures. Communications of the
ACM, Vol. 15, pp. 11–15.
Falco F. D. and Mele R. 2002, The monitoring of bridges for scour by sonar and sedimentri. NDT & E International, Vol. 35, pp.
117–123.
Firat M. 2009, Scour depth prediction at bridge piers by Anfis approach. Proceedings of ICE-Water Management, Vol. 162, No. 4,
pp. 279–288.
Forde M. C., McCann D. M., Clark M. R., Broughton K. J., Fenning P. J., and Brown A. 1999, Radar measurement of bridge scour.
NDT & E International, Vol. 32, pp. 481–492.
Fukui J. and Otuka M. 2002, Development of new inspection method on scour condition around existing bridge foundations.
Proceedings of First International Conference on Scour of Foundation, Texas, USA, pp. 410–420.
Gonzalez R. C. and Woods R. E. 2002, Digital Image Processing. 2nd Edition, Prentice Hall, New Jersey, USA.
Guven A., Azamathulla H. Md., and Gunal M. 2012, Comparative study of predicting scour around a circular pile. ICE-Maritime
Engineering, Vol. 165, No. 1, pp. 31–40.
Haralick R. M., Sternberg S. R., and Zhuang X. 1987, Image analysis using mathematical morphology. IEEE Transactions on Pattern
Analysis and Machine Intelligence, Vol. PAMI-9, No. 4, pp. 532–550.
Khan M., Azamathulla H. Md., Tufail M., and Ghani A. Ab. 2012a, Bridge pier scour prediction by Gene-expression Programming,
ICE-Water Management, Vol. 165, No. 9, pp. 481–493.
Khan M., Azamathulla H. Md., and Tufail M. 2012b, Gene-expression programming to predict Pier scour depth using Laboratory
data. IWA Journal of Hydroinformatics, Vol. 14, No. 3, pp. 628–645.
Kothyari U. C., Garde R. J., and Range Raju K. G. 1992, Temporal variation of scour around circular bridge piers. Journal of
Hydraulic Engineering, Vol. 118, No. 8, pp. 1091–1106.
Lin Y. B., Chen J. C., Chang K. C., Chern J. C., and Lai J. S. 2005, Real-Time Monitoring of Local Scour by Using Fiber Bragg
Grating Sensors. Smart Materials & Structures, Vol. 14, No. 4, pp. 664–670.
Lin Y. B, Lai J. S., Chang K. C., Chang W. Y., Lee F. T., and Tan Y. C. 2010, Using MEMS Sensors in the Bridge Scour Monitoring
System. Journal of the Chinese Institute of Engineers, Vol. 33, No. 1, pp. 25–35.
- 276 - International Journal of Sediment Research, Vol. 29, No. 2, 2014, pp. 269–277
Lopes A. C. and Dias J. M. S. 2006, Integration of Geo-Referenced Data for Visual Simulation in Location-Based Mobile
Computing. Computer-Aided Civil and Infrastructure Engineering, Vol. 21, pp. 514–529.
Lu J. Y., Hong J. H., Su C. C., Wang C. Y., and Lai J. S. 2008a, Field measurements and simulation of bridge scour depth variations
during floods. Journal of Hydraulic Engineering, Vol. 134, No. 6, pp. 810–821.
Lu K. H., Wang C. M., and Chen S. Y. 2008b, Traffic Light Recognition. Journal of the Chinese Institute of Engineers, Vol. 31, No.
6, pp. 1069–1075.
Melville B. W. 1997, Pier and abutment scour—an integrated approach. Journal of Hydraulic Engineering, Vol. 123, No. 2, pp. 125–
136.
Melville B. W. and Chiew Y. M. 1999, Time scale for local scour at bridge piers. Journal of Hydraulic Engineering, Vol. 125, No.1,
pp. 59–65.
Melville B. W. and Coleman S. E. 2000, Bridge Scour. Water Resources Publications, Highland Ranch, Colorado, USA.
Mia F. and Nago H. 2003, Design Method of Time-Dependent Local Scour at Circular Bridge Pier. Journal of Hydraulic
Engineering, Vol. 129, No.6, pp. 420–427.
Oliveto G. and Hager W. H. 2002, Temporal Evolution of Clear-Water Pier and Abutment Scour. Journal of Hydraulic Engineering,
Vol. 128, No. 9, pp. 97–105.
Oliveto G. and Hager W. H. 2005, Further results to time-dependent local scour at bridge element. Journal of Hydraulic Engineering,
Vol. 131, No. 2, pp. 811–820.
Otsu, N. 1979, A threshold selection method from gray-level histograms. IEEE Transactions on Systems, Man, and Cybernetics—
Part B: Cybernetics, Vol. 9, pp. 62–66.
Raudkivi A. J. and Ettema R. 1983, Clear-water scour at cylindrical piers. Journal of Hydraulic Engineering ASCE, Vol. 109, No. 3,
pp. 338–350.
Shen H. W., Schneider V. R., and Karaki S. S. 1969, Local scour around bridge piers. Journal of Hydraulic Engineering ASCE, Vol.
95, No. 6, pp. 1919–1940.
Wang H., Wang C. Y., Lin C., Liu H. C., Hu C. H., and Chen M. H. 2010, A Forensic Study on the Collapse of Shuang-Yuan Bridge
during Typhoon Morakot in Taiwan. Proceedings of the 5th Civil Engineering Conference in the Asian Region (CECAR5) and
Australasian Structural Engineering Conference 2010, Sydney, Australia, Paper ID # 0525.
Yankielun N. E. and Zabilansky L. 1999, Laboratory Investigation of Time-domain Reflectometry System for Monitoring Bridge
Scour. Journal of Hydraulic Engineering, Vol. 125, No. 12, pp. 1279–1284.

International Journal of Sediment Research, Vol. 29, No. 2, 2014, pp. 269–277 - 277 -