Automation in Construction 16 (2007) 546 – 555

www.elsevier.com/locate/autcon

Innovative process to characterize buried utilities using

Ground Penetrating Radar
Jim Lester a , Leonhard E. Bernold b,⁎
a
Control Systems Engineer, E-Merge Systems Inc., Richmond, VA, 23220, USA
b
Department of Civil Engineering, NC State University, Raleigh, NC 27695, USA
Accepted 15 September 2006

Abstract

Today's non-invasive technologies for locating buried utilities can be considered as ancient. However, Ground Penetrating Radar (GPR) has
recently received significant attention from the scientific community since it showed great promise in detecting landmines. Yet, the complexities of
the underground, especially in inhabited areas, makes “seeing-through-the-earth” to find buried utilities extremely difficult. This paper presents the
results of a data processing method, called Translation Invariant Wavelet Packet Detection (TIWPD), applied to filtering GPR data collected on a
university campus. It first provides a brief introduction into the working principles of scanning the ground with electromagnetic radar waves that
are being refracted, scattered, and reflected by buried objects of all sizes and materials. In its main section, the paper presents the results of
experimental deployment of the system during a construction project that involved the extensive excavation trenches to lay chilled water pipes.
The significance of this paper lies in its use of real-world GPR data to demonstrate the performance characteristics of the filtering process and its
validation with the actual condition found during excavation. The encouraging results of this work should provide the basis for developing a near-
real time utility detection system that can be used by laborers in the field.
Published by Elsevier B.V.

Keywords: Ground Penetrating Radar; Data filtering; Buried utility protection; Non-invasive utility detection; Accident prevention; Translation Invariant Wavelet
Packet Detection

1. Introduction and background nications, water supply, electric power, and other vital public
services, such as hospital and air traffic control operations, and
The demand for new buried utilities is growing with new is a leading cause of natural gas and hazardous liquid pipeline
construction, re-construction, and the growth of the subsurface accidents.” In most any state of the U.S. a contractor is required
infrastructure worldwide. As a result, contractors are busy by law to call a “One-Call Center” 48 or more hours before
digging and trenching into the ground in order to bury new digging. Connecticut was one of the first states to pass a “call
pipes. Because the machinery for placing the new utilities Before you Dig Law” in 1978 which had the effect that
underground, such as backhoe excavators, trenchers, augers, accidental cuts of utility lines declined by 60% while the length
drills, and plows, don't “see” when they are getting close to of new underground utilities being buried increased. The centers
already buried objects, utilities are easily damaged. serve as the clearinghouse for excavation activities that are
The congressional Transportation Equity Act for the 21st planned close to pipelines and other underground utilities. “Our
Century, TEA 21, Title VII, Subtitle C, SEC. 87301, states that: purpose is to prevent damage to underground facilities. To
“…unintentional damage to underground facilities during promote safety, we provide an efficient, cost effective
excavation is a significant cause of disruptions in telecommu- communications network among designers, excavators, and
facility owners.” (http://www.pipeline101.com/PipelinesYou/
⁎ Corresponding author. Tel.: +1 919 515 3677. onecall.html.) A key problem of the presently available hand-
E-mail addresses: [email protected] (J. Lester), [email protected] held technologies to detect underground utilities before digging
(L.E. Bernold). is the inability to reliably predict the exact location of existing
0926-5805/$ - see front matter. Published by Elsevier B.V.
doi:10.1016/j.autcon.2006.09.004
 J. Lester, L.E. Bernold / Automation in Construction 16 (2007) 546–555 547

utilities. As Fig. 1 shows the results of a damaged gas utility Table 1

which cost many lives every year and are responsible for large Dielectric constants and wave velocities of common subsurface media
consequential damages [1]. Dielectric Constant εr Velocity (m/ns)
For locating of buried utilities a variety of tools are used today Air 1 0.3
but there exists not one technology that offers satisfying results Water 81 0.033
for the various types of utilities found in the ground. The Com- Rocks 4–12 0.15–0.087
Sand dry 3–5 0.15–0.12
mission on Geosciences, Environment and Resources (CGER)
Sand wet 20–30 0.055
of the National Academy of Sciences stressed this point when it Clay dry 2–6 0.11–0.09
lamented [2]:“Much of the site characterization conducted today Clay wet 15–40 0.052
uses techniques that are more than 20 years old. … Noninvasive Concrete 9–12 0.10–0.087
site characterization would probably be used more frequently Asphalt 3–6 0.17–0.12
and efficiently if much of the data acquisition, data processing,
and decision-making could be automated. … By producing a place since 1929 when it was used to determine glacier depth,
better result, more rapidly and at lower cost, robotics and de- but since the 1970's it has evolved to be used for applications
cision support system could be the key to more–and more such as archaeology, detection of landmines, forensics, and
effective–use of noninvasive site characterization methods. mapping underground utilities. The majority of today's GPR
Automation of site characterization allows measurements and technology is based on ultra-wideband (UWB) impulse radar
preliminary interpretation to be made in real time.” principles due to the depths at which most targets are located
In 1993, the Construction Automation and Robotics Labora- [3]. The velocity of electromagnetic waves propagating through
tory (CARL) at NC State University started an initiative to address a media is a function of its dielectrics or its capacity to work as
the problem of finding ways to protect utility during construction. an insulator. In rocks and sediments, dielectric properties are
The core idea was to develop a tool that lets equipment operators primarily a function of mineralogy, porosity, water saturation,
sense the existence of a buried utility in the vicinity of the cutting frequency, and depend on the rock lithology, component
tool while the equipment is in use. While mounted on the geometries, and electrochemical interactions [4]. In addition,
machine, such a sensing device had to be able to “see through” the researchers were able to show that the dielectric constant of soil
soil and reliably warn the operator. This paper discusses the work is a function of the percent of water in the soil, increasing with
that was done to improve the state-of-the-art in processing sensory increasing water % in the soil [5]. The combined effect of all the
data generated by a Ground Penetrating Radar (GPR). properties are known as dielectric constant, a material constant
which controls the speed of electromagnetic waves in the
2. Seeing through the ground with Ground Penetrating material. Table 1 presents the dielectric constants (dimension-
Radar less) and velocities of electromagnetic waves through various
media found underground.
The Ground Penetrating Radar (GPR) is a rapid, high- The data in Table 1 highlights three important features: 1) The
resolution tool for non-invasive investigation of the ground that harder the material the slower the velocity, 2) the higher the water
sends out microwave radiation that passes through the ground content the slower the velocity, and 3) the velocity in a material
and records what is reflected back to the surface. It has been in can vary significantly. As will be discussed in the following
sections, the variations in velocities make it extremely hard to
accurately predict the depth of a buried object.

Fig. 1. Running engine ignites escaping gas after dozer ripped a buried pipe. Fig. 2. Schematic of UWB impulse GPR unit and diffraction mechanism.
548 J. Lester, L.E. Bernold / Automation in Construction 16 (2007) 546–555

Fig. 4 b) presents an easy to recognize feature in a GPR scan that

could have been created by a variety of buried objects that did not
contain metal pipes, since they display a unique pattern. The GPR
scan had been collected using the mobile utility detection cart
shown in Fig. 4 a) prior to trenching for a new water pipe.
According to the available as-built drawings nothing should have
been underground at the location. As a consequence, the contractor
continued his operation as planned only to find the concrete duct
bank shown in Fig. 4 c). Getting around (underneath) the obstacles
resulted in extra work that had not been planned for.
While unique features are readily recognizable by an
experienced user of GPR smaller size objects within a messy
subsurface are extremely difficult to discern, even by experts.
As a consequence it is desirable to remove, or filter, the effect of
Fig. 3. Three views of radar trace profiles in homogenous material. clutter on a GPR data set. The term clutter, often referred to as
background, refers to any portion of a GPR trace that is
2.1. Basics of impulse radar unrelated to a target. The use of clutter suppression techniques
allows persons who have never before worked with GPR data to
Fig. 2 presents the key elements of a GPR unit: 1) Pulse identify targets that would normally be obscured. Both
Generator, 2) transmitter, 3) receiver antenna, and 4) processing continuous and discrete wavelet analysis have found wide-
circuitry. The main component of the Pulse Generator is a spread use in studying fluid turbulences or the propagation of
power source capable of transmitting an electrical signal with ultrasonic waves through material [6]. The usefulness of the
high peak voltage and short time duration. This signal is used to Discrete Wavelet Transform (DWT) applied to ultrasonic
produce a non-sinusoidal electromagnetic (EM) pulse, which testing has been demonstrated in several instances [7,8]. In
can be represented as a Gaussian waveform. The energy from particular, DWT is being used to improve the reliability of land-
the EM pulse is radiated outward in straight lines from the or antipersonnel mine detection systems [9].
Transmit Antenna and reflected back towards the receive Because of their widespread usage, the clutter suppression
antenna by any media and/or targets in the transmission path. methods of mean background trace subtraction (MBTS), and
Fig. 2 also highlights the three main scenarios of wave matched filter deconvolution (MFD), will be compared with 3)
behavior. While certain pulses are being reflected back to the Translation Invariant Wavelet Packet Detection (TIWPD).
receiver, others are being scattered and refracted. The echo While MBTS and MFD are used to simply enhance the
captured by the receiver is called a trace and represents the visibility of targets, TIWPD is beginning to be used to clean up
cumulative reflections over a short period of time (e.g., the clutter of GPR images by nullifying target-free GPR traces.
nanoseconds) in time after a pulse was launched from the
transmitter. Fig. 3 presents the result of moving the GPR across a 3. Common GPR signal processing methods
homogenous half-space with a constant material property.
However, the world's underground is seldom homogenous Over the course of the last decade a variety of signal
especially in areas that have been “transformed” by humans. processing routines have been developed and tailored for use
Layers of sands, clays, gravel, are mixed with stones, tree roots, with GPR. Zoubir et al. [10] provide a review of the MBTS,
construction debris, and of course, utilities of all kinds. The MFD, trimmed average power, TIWPD, and Kalman filtering.
resulting radar images show streaks of black and white bands in The goal of all these methods is to suppress background noise
various forms and shapes, magically split and joined at all angles. (or clutter) since it obscures a target's visibility.

Fig. 4. Field experiment with GPR.

 J. Lester, L.E. Bernold / Automation in Construction 16 (2007) 546–555 549

3.1. The mean background trace subtraction filter

One of the most common filters used for GPR data is called
the mean background trace subtraction (MBTS) filter, modeled
in Fig. 5.
The equation is of the form:
Fig. 6. Block diagram of deconvolution operation.
xðt Þ ¼ nd sðt Þ þ b8
ðt Þ; for−l < t
< sðt Þ ¼ target for n ¼ 1 the MBTS method, then from Eq. (3) MFD can be used to
< l where sðt Þ ¼ target  free for ni0 ð1Þ recover m(t).
:
bðt Þ ¼ background noise
xðt Þ ¼ vðt Þ* ĥ ðtÞZ vðt Þ ¼ deconvolveðxðt Þ; ĥðt ÞÞ
The key to the success of this method is that there exists a slow deconvolveðxðt Þ; ĥðt ÞÞ−dðt Þ¼ mðt Þ; where mðt Þis a target for n ¼ 1
spatial variation of the current soil conditions. Under this con- deconvolveðxðt Þ; ĥðt ÞÞ−dðt Þc0; where mðtÞis target  free for nc0
dition it is possible to calculate a mean background trace (b̂ (t))
ð4Þ
using a moving average window of N GPR traces, and by
substituting b̂ (t) into Eq. (1) it is possible to greatly enhance target Again, much like the MBTS method, it is important to ensure
visibility by subtracting the mean background trace. that the window size N used to calculate hˆ(t) is neither too
X
kþN wide, nor too narrow. One of the inherent problems in using a
bk ðt Þ moving average (MA) window to dampen the effect of
xðt Þ− b̂ðt Þ ¼ sðt Þ
xðt Þ− b̂ðt Þc0 g where b̂ðt Þ ¼ i¼k
N
ð2Þ background signals is that useful target information is also
used in the MA calculation. For targets at the far end of the
The successful implementation of MBTS hinges on the effective GPR detection range, whose signature is already weak,
realization that the total window size N must be wide enough to the MA methods can essentially erase their presence from the
produce an accurate b̂ (t), while narrow enough to account for processed data set. Therefore, a method has been developed that
local changes in the soil conditions. relies on the detection of abrupt changes in the energy contained
in a GPR trace to signify the presence of a target.

3.2. The matched filter deconvolution filter 3.3. The Translation Invariant Wavelet Packet Detection filter

Another method that utilizes a moving average window of Valuable information from the time domain is lost when a signal
background traces, but for a different reason, is known as the is transformed via Fourier methods to the frequency domain.
matched filter deconvolution. A convolution operation, referred to Fourier analysis offers no way of analyzing abrupt changes, such as
as the filter, is able to remove noise or other unwanted components those that are produced by a target in a GPR trace. Wavelet
from an output signal x(t) as the model in Fig. 6 depicts. analysis, on the other hand, is capable of handling data sets
The analytical equation of this operation is of the form: containing abrupt changes. Wavelets are mathematical functions
that break a given signal into different frequency components thus
creating the opportunity to cut out and study “windows” of data
xðt Þ ¼ ½dðt Þ þ nd mðt Þ*h8ðt Þ
according to scales or resolutions. As a result, large and small
> dðt Þ ¼ GPR impulse function
>
>
> features within a signal can be studied separately and provides a
>
> mðt Þ ¼ target for n ¼ 1
< critical advantage in cases where signals show sharp changes
mðt Þ ¼ target  free for ni0
¼ vðt Þ*hðt Þ; where ð3Þ within, like GPR pulses. Shin and Hryciw [11] presented a
>
> v ðt Þ ¼ ½dðt Þ þ nd mðt Þ
>
> promising application of wavelet transform analysis to determine,
>
> h ðt Þ ¼ background noise
: from black and white images, the average grain size of soil with
* ¼ convolution operator
pixels per diameter between 1 and 50. The discrete DWT analysis,
mentioned earlier, is a procedure by which a discretely sampled
If an estimate of the background signal h(t), referred to as signal is decomposed into two sets resulting in an approximation
ĥ(t), is calculated using the same general approach taken in coefficient (A), representing the low frequency components, and a
detail coefficient (D) representing high frequency components.
Each approximation can be further split. Wavelet packet analysis is
a procedure that allows both the approximations (A) and the details
(D) to be split. This will form a wavelet packet decomposition tree
like the one shown in Fig. 7 [12]. The original signal s is
N−1
decomposed into 22 representations and can be reconstructed,
or synthesized, by D1 +DA2 + AAA3 +DAA3. Another representa-
tion of the signal might be AA2 +DA2 +D1. In order to find the best
Fig. 5. Representation of superposition equation for the received GPR signal x(t). representation of the original signal the entropy-based cost
550 J. Lester, L.E. Bernold / Automation in Construction 16 (2007) 546–555

Fig. 7. Wavelet packet decomposition tree of level N = 3.

function, E(s) is used. For this paper a cost function that computes Once the best basis is computed, a test statistic (TS) is calculated
the log of the energy (Eq. (5)) contained in a given set of wavelet from the wavelet packet coefficients and compared to an
packets was used to compute the signal's best representation. adaptively changing background signal test statistic (T0). A
X   significant increase in TS occurs when a target is encountered
E ðsÞ ¼ log s2i ð5Þ and the threshold defined by T0 is exceeded. When this happens
i T0 is no longer updated until TS falls below a minimum
Wavelet packet filtering seems to be an effective method to threshold (T1). All traces that fall outside of the hysteresis
analyze GPR traces due to its ability to adapt to the changing formed by T0 and T1 are considered to be clutter and are
time, scale, and frequency contents of background signals subsequently set to 0.
produced by soil conditions and targets. However, as the GPR When using the TIWPD method it is important that the first
unit is along the surface variations in the position of the receiver N traces of sampled data be target free allowing for an accurate
antenna relative to the target produce time shifts in the data sets. initialization of the T0 and T1 hysteresis. For the analyses
According to Carevic [13], this shortcoming has been over- presented in this paper N will be set to 20. The remaining part of
come with Translation Invariant Wavelet Packet Detection the paper will be used to compare the effectiveness of the three
(TIWPD). This method computes the best basis of two shifts discussed filtering methods, MBTS, MFD, and TIWPD, in
of the trace by minimizing the entropy based cost function. identifying the existence of buried utilities.

Fig. 8. Examples of the congested underground space at a university campus.

 J. Lester, L.E. Bernold / Automation in Construction 16 (2007) 546–555 551

4. Comparative evaluation of data processing filters Due to the recognized importance of improving the technolo-
gies for locating underground utilities all of the involved agencies
Only small segments of North Carolina consist of a sandy and companies agreed to let us scan the subsurface before
subsurface while the predominant media is clay making it the excavation and identify the type, size, and the exact location of the
largest brick-producing state in the Nation. Since interpreting exposed utilities after the trenching was complete. Buried pipes
GPR data collected in clayey soils has been found very difficult, are made of a variety of material such as bore steel, ductile iron,
the comparative evaluation was applied to this type of material. terracotta, plastic, and concrete. The size varies from 1 in. (2.5 cm)
to 4 ft (122 cm) diameters. In addition, one can find power cables
4.1. Collecting GPR data in clay type material of all kinds some still inside their decayed wooden ducts. A
constant source of false alarms are the many abandoned lines that
In addition to test GPR in clay it was found necessary to use nobody knew about but lay next to or over hot lines. As
data from an urban setting where the underground has been mentioned before, the as-built drawings had not been updated and
previously disturbed. For this purpose, the research team thus were of little help.
collaborated with a pipe-laying contractor working on a new Prior to the start of excavating a section, the research team
Chilled Water plant at NC State University that required the scanned it with a MALA Geoscience RAMAC/GPR with a
installation of a pipe system to circulate the water between the shielded 500 MHz antenna mounted on a mobile cart that also
many buildings and the plant. This excavation effort provided an measured the travel distance using a wheel encoder. It is interesting
excellent opportunity to collect real world data especially since to note that, as the research team became more effective in
the subsurface of a college campus contains a complex network interpreting the GPR data, the excavator foreman started to request
of utilities both in use and abandoned. Figs. 8 and 9 shows that the team to mark their predictions on the asphalt. The data
several examples of an overcrowded subsurface that many refer collected during each scan, GPR traces and traveling distance, was
to as the “Wild-West of the 21 century”. stored on a laptop and processed in the office using the three

Fig. 9. Investigation of a congested subsurface section.

552 J. Lester, L.E. Bernold / Automation in Construction 16 (2007) 546–555

Fig. 10. GPR data after TIWPD filtering.

filtering methods. As the excavating crew opened the trenches to depths. The utilities labeled C and D were destroyed during
visual inspection the unearthed utilities were photographed and excavation but were drawn into the photograph for clarity.
their size and burial depth recorded. Unfortunately, utilities that had Indicated in the Fig. 9 a) and b) is the direction of the GPR scan
been previously abandoned were destroyed in the process but the which started approximately 1.8 m before the buried duct bank
excavator operator kept pieces of the destroyed utility for the at the edge of a concrete walkway.
research team to examine. A comparison of the GPR data collected on the beach, shown
in Fig. 4 b), with the plot from the on-campus scan underscores
4.2. Locating utilities buried in clay the drastic effect of the non-homogenous subsurface. Without
the a priori knowledge of what is buried where, it would be
The GPR data sample used for this analysis was generated extremely difficult to recognize the presence of particular
during scan behind Withers Hall. As the open trench in Fig. 9 a) objects. The same holds true after the raw data is filtered using
illustrates, the selected 10 m contained a multitude of buried the MFD and the MBTS methods both depicted in a 3-D view in
objects among them two terracotta pipes (one not shown in Fig. 9 c) and d). The usefulness of the MBTS and MFD methods
picture), a concrete duct bank, and various pipe sizes at various lies in their ability to reduce background noise by averaging N

Fig. 11. Crisp utility recognition with the TIWPD method.

 J. Lester, L.E. Bernold / Automation in Construction 16 (2007) 546–555 553

traces. However, the removing of background noise also lessens unmistakably identifiable. On the other hand, the two pipes
the already weak target signatures. As a consequence, duct bank crossing the trench in a non-perpendicular angle create features
as well as pipe A are impossible to be determined with accuracy. that need to be explored in more detail. As Fig. 9 a) indicates,
This situation changes drastically with the use of the TIWPD “cylinder” D is made up of two steel pipes encased in concrete
filtering method. The difference in the meaningfulness of the inside a concrete pipe. Since the TIWPD method focuses on
data plots shown in Fig. 10 is self-evident. Both, in 2-D and the detecting abrupt changes in signal strength through the use of
3-D, the graphs clearly indicate the location as well as the the test statistics (T0 and T1) it is reasonable to expect that
boundaries of the objects. Most surprisingly, even the clay-pipe separate spikes result when the distance between the instances
underneath the concrete walkway was clearly identified. are large enough for new traces to fall bellow T1. In the case of
Similarly, pipe A shows up separate from the duct-bank while pipe D, an imaginary scan “cuts” in a 45° angle through the
the cables and pipes inside the duct, which could not be cylinder creating an oval cutting surface slicing through four
inspected, left their fingerprints on both plots. Fig. 10 b) gives drastically different media: a) Concrete casing, b) metal pipe 1,
also an indication about the approximate bottom of the concrete c) concrete center, d) metal pipe 2, and e) concrete casing. With
duct bank which could be turned into meters and feet if the this in mind it can be inferred that the first peak is triggered by
velocity of the electromagnetic wave passing through clay the drastic change in the dielectric constants wet clay (εr = 30)
would be known. Unfortunately, its accurate determination is and of concrete (εr = 10). The magnetic steel of the two pipes is
non-trivial and changes according to the characteristics of the causing another large discontinuity that is being recognized and
soil, in particular its water content. The simple rule of thumb repeated as the GPR crosses over the center of the encased set of
says: The less water the higher the velocity (for the same soil). pipes. TIWPD's effectiveness in detecting changes in the
Based on the demonstrated effectiveness of the TIWPD underground holds also true at the “backside” of a target in that
filtering method the remainder of the paper is used to present its a sharp decrease in signal strength, produced by its trailing edge,
results when applied to a variety of different situations found in also triggers the TIWPD test statistics. This very trait is also
the underground on NCSU's campus. Fig. 11 features the being produced by object C. Here, the pipe consists of non-
results of filtering the GPR data collected in the 6 m stretch after metallic material, terracotta, and is encased by concrete. In the
the concrete duct shown in Fig. 9 a). As the two plots in Fig. 11 absence of a metal core only two drastic changes emerge, each
illustrate, the four buried objects A, B, C, and D are time between the surrounding clay and the concrete. The latter

Fig. 12. GPR features of a buried concrete drainage pipe inside backfilled material.
554 J. Lester, L.E. Bernold / Automation in Construction 16 (2007) 546–555

explanation has to be hypothetical since the backhoe excavator this point T0 is surpassed twice at E and F. While the object that
destroyed the pipe. triggered E is hard to make out, could be a metal rod/cable, F
The fact that the TIWPD filter so accurately identified indicates the position of the second trench-wall.
terracotta pipes is an extremely valuable quality since many
utilities are non-ferrous and thus don't transmit induced radio 5. Significance of the research results
signals traceable from the surface. Because of its importance the
filter's performance in locating larger non-metallic drainage As mentioned earlier, the Commission on Geosciences,
pipes needs to be studied. Again, the chilled water project Environment and Resources [2] stressed the need for new
offered an excellent opportunity to compare the GPR output technologies to drastically improve our capabilities to see what
with the actual situation found during trenching. is underground. The importance of this call for protecting the
subsurface infrastructure can be found in many publications.
4.3. Detecting large tubes underground For example, Sterling [14] surmised in his report that we need
“techniques that accurately resolve the position and type of an
Fig. 12 presents the situation and the GPR reading of a 0.6 m underground utility in the presence of other underground util-
(2 ft) diameter concrete drainage pipe crossing the street at a ities and structures.” Jeong et al., [15] lamented about the
depth of 0.8 m (32 in.). The TIWPD plot in Fig. 12 b) exposes drawbacks of current methods such as GPR with its “inapplic-
several empty sections while revealing the 1.8 m (6 ft) stretch ability to high conductive soils, clay and saturated soil, practical
between A and D as a “foreign” object with a contour not limitation of imaging objects located 2 m below the surface,…”
resembling that of a concrete duct bank as in Fig. 10 a). Lining On the other hand Jeong et al. also found in a survey that the
up the raw with the filtered GPR data while taking into account most significant issue is “the unavailability of adequate infor-
the method of construction it is feasible to interpreting the mation for existing underground utilities (that) causes problems
mysterious alignment of the traces. in searching and finding surface appurtenances (starting point
The laying of large concrete pipes is commonly done by of utility tracing) and selecting appropriate equipment for
digging a trench that is wider than the pipe diameter in order for tracing utilities” [15]. Different research groups are working
the laborers to stand and, in addition, to allow for the required on various aspects of the problem. For example Witten
haunch width. After the proper depth is excavated, crushed Technologies [16] integrated several GPR units to switch from
stone for the bedding is laid down, the pipe is aligned, followed a “one slice” to a spatially integrated view of the underground.
by proper backfilling with compaction. In Fig. 12 b) the spatial A team sponsored by Gas Technology Institute [17] is working
contour of the trench wall, bedding, and haunch are drawn on on mounting an obstacle detection sensor on the head of a
top of the raw GPR data. As the GPR reaches position marker A horizontal directional drilling equipment. A similar approach,
its first electromagnetic wave, emitted in a cone shaped pattern, adding a sensor to the digging equipment, has been success-
are interfering with the old trench wall and the still existing fully tested by Bopana and Bernold [18]. The work presented
crushed stone bedding underneath the pipe. As the GPR moves in this paper adds three new components: 1) A data processing
closer to the pipe, more and more of its waves interfere with the method that drastically improves the characterization capabil-
trench bottom and backscatter a larger and larger quantity. At ities of a GPR, 2) use of the discrete wave analysis method that
the same time, the effect of the old trench wall becomes more enables it to be real-time, thus eliminating the commonly
apparent. This is caused by the change in soil material needed time for post-processing the large amount of data,
characteristics especially the water content. The difference in and 3) thorough field testing of the system with data
dielectric constants between the natural and the backfill material verification during the trenching operation.
will refracted depending on the differences in dielectric The next step to further this approach that has never been
constants. Changes in refraction as well as amount of deflection used in utility detection is to automate the locating process by
are not limited to the trench wall but also the round contour of adding a feature recognition capability. In addition, the GPR
the concrete pipe, haunch, bedding, and the bottom of the trench will be integrated with an electromagnetic sensor to more easily
excavation. It goes without saying that the resulting mesh of differentiate between metallic and non-metallic objects.
“crisscrossing” waves are impossible to model without the
detail knowledge of what material had been used, its present 6. Summary and conclusions
water content, and exact spatial dimensions.
As the TIWPD filter considers the change in energy that is In order to better protect buried utilities better tools are
reflected between the locations A and D, the traces are being needed to accurately locate them. One such tool is Ground
pasted from the raw GPR onto its output plot. The presented top Penetrating Radar (GPR). This paper introduced and compared
view of those traces show two distinct “ridges”, above the three filtering methods to process the traces generated by a
trench wall at location B and above the beginning of the pipe at GPR. First, the basics of the GPR technology were introduced
C. On the other hand, the backscattered waves lose intensity followed by a discussion of the many factors that effect data
between C and D, which is due the intense refraction and interpretation and three signal processing methods that could
scattering effect in between the trench wall and the rounded assist in the processing of the data. The Translation Invariant
pipe. Above the center of the pipe, at location D, the energy of Wavelet Packet Detection (TIWPD) method showed excellent
the trace is smaller than T1 turning the object flag to off. Beyond results. Real field data, collected ahead of an excavation crew,
 J. Lester, L.E. Bernold / Automation in Construction 16 (2007) 546–555 555

was filtered with the TIWPD and its output compared with the [5] A. Martinez, A.P. Byrnes, Modeling Dielectric-Constant values of Geologic
actual situation that was uncovered during the trenching Materials: An Aid to Ground-Penetrating Radar Data Collection and Inter-
pretation, Current Research in Earth Sciences, Bulletin, vol. 247, Aug. 2005,
operation. For two locations the raw and processed data was part 1. http://www.kgs.ukans.edu/Current/2001/martinez/martinez1.hmtl.
overlaid with the real-world situation. Using the understanding [6] F. Lanza di Scalea, J. McNamara, Measuring high-frequency wave
of GPR technology and construction methods it was possible to propagation in railroad tracks by joint time frequency analysis, J. Sound
demonstrate the effectiveness of the TIWPD filter in pin- Vib. 273 (2004) 637–651.
[7] P. Addison, K. Murray, J.N. Watson, Wavelet transform analysis of open
pointing the location of ferrous and non-ferrous utilities. The
channel wake flows, J. Eng. Mech., ASCE 127 (1) (2001) 58–70.
relatively small computational load required to implement the [8] P. Rizzo, F. Lanza di Scalea, S. Banerjee, A. Mal, Ultrasonic characterization
TIWPD method, as well as its simplistic target visualization and inspection of open cell foams, J. Eng. Mech., ASCE 131 (11) (2005)
method, makes the TIWPD method a promising candidate for 1200–1208.
use by excavation crews. [9] S. Perrin, E. Duflos, P. Vanheeghe, A. Bibaut, Multisensor fusion in the
frame of evidence theory for landmines detection, systems, man, and
cybernetics, IEEE Transaction 34 (4) (2004) 485–498.
Acknowledgements [10] A.M. Zoubir, I.J. Chant, C.L. Brown, B. Barkat, C. Abeynayake, Signal
processing techniques for landmine detection using impulse ground
Funding for the work has been provided by Public Health penetrating radar, IEEE Sens. J. 2 (1) (2002) 41–51.
Service-National Institutes of Health under contract 5 R01 [11] S. Shin, R.D. Hryciw, Wavelet analysis of soil images for particle size
CCR413051–02 and 1R01 0H04201–01. Its content is solely determination, J. Comput. Civ. Eng., ASCE 18 (1) (2004) 19–27.
[12] M. Misiti, Y. Misiti, G. Oppenheim, J.-M. Poggi, Wavelets: A New Tool
the responsibility of the writers and do not necessarily represent for Signal Analysis, MATLAB Release 13 — Wavelet Toolbox
the official views of NIH. The authors also acknowledge the Documentation, The MathWorks, Inc., Natick, Massachusetts, 2002.
contributions made by the many members of the BUDS team [13] D. Carevic, Clutter reduction and target detection in ground penetrating
working at the Construction Automation & Robotics Laboratory. radar data using wavelets, SPIE Conf. Detection Remediation Technologies
The work would not have been possible without the cooperation Mines, Minelike Targets IV, SPIE, Orlando, Florida, 1999, pp. 973–978.
[14] R. Sterling, Utility locating technologies: a summary of responses to a
of the Construction Management Department of the NC State statement of need distributed by the Federal Laboratory Consortium for
University, Gilbane Building Company managing the project, and Technology Transfer, Federal Laboratory Consortium Special Reports
the MIS Contracting Company which installed the new pipes. Series No. 9, Louisiana Tech University, LA, 2000.
[15] H.S. Jeong, D.M. Abraham, J.J. Lew, Evaluation of an emerging market in
References subsurface utility engineering, J. Constr. Eng. Manag., ASCE 130 (2)
(2004) 225–234.
[16] R. Birken, D.E. Miller, M. Burns, P. Albats, R. Casadonte, R. Deming, T.
[1] L.E. Bernold, Economic model to optimize underground utility protection, Derubeis, T. Hansen, M. Oristaglio, Efficient Large-Scale Underground Utility
J. Constr. Eng. Manag., ASCE 12 (6) (2003) 645–652. Mapping Using a New Multi-Channel Ground-Penetrating Imaging Radar
[2] Commission on Geosciences, Environment and Resources, Seeing Into
System, Witten Technologies Inc., Boston, MA., August 31 2006 <http://www.
The Earth: Noninvasive Characterization of the Shallow Subsurface for dot.ca.gov/hq/esc/geotech/gg/geophysics2002/042birken_gpir.pdf >.
Environmental and Engineering Applications, National Academy Press, [17] M.J. Kieba, C.J. Ziolkowski, Differential Soil Impedance Obstacle
Washington, DC, 2002.
Detection, Quarterly Report, Gas Technology Institute, Des Plaines,
[3] D.J. Daniels, Surface-Penetrating Radar, The Institution of Electrical
Illinois, 2004, August 31 2006 <http://www.osti.gov/bridge/servlets/purl/
Engineers, London, 1996. 834183-lc1y6Y/native/834183.pdf >.
[4] F.M. Francisca, V.A. Rinaldi, Complex dielectric permittivity of soil– [18] B.T. Kolera, L.E. Bernold, An intelligent utility locating tool for excavators,
organic mixtures — 20 MHz–1.3 GHz, J. Environ. Eng. ASCE 129 (4)
J. Constr. Eng. Manag., ASCE 132 (9) (2006) 919–927.
(2003) 347–357.