Automation in Construction 73 (2017) 111

Contents lists available at ScienceDirect

Automation in Construction

journal homepage: www.elsevier.com/locate/autcon

A technology for sewer pipe inspection (Part 2): Experimental

assessment of a new laser proler for sewer defect detection
and quantication
Mathieu Lepot a,, Nikola Stani a, Franois H.L.R. Clemens a,b
a
Delft University of Technology, PO Box 5048, 2600, GA, Delft, The Netherlands
b
Deltares, PO Box 177, 2600, MH, Delft, The Netherlands

a r t i c l e i n f o a b s t r a c t

Article history: In order to test the laser-scanning device presented by Stani, Lepot, Catieau, Langeveld and Clemens [1], labora-
Received 28 August 2015 tory experiments have been performed. Various objects, covering a wide range of sizes, shapes, materials, colours
Received in revised form 25 October 2016 and reectiveness, have been installed in a concrete pipe and scanned by the prototype in order to identify po-
Accepted 31 October 2016
tential object characteristics that may affect measurement uncertainty and/or create bias. By taking into account
Available online xxxx
both uncertainties (on scanned and measured sizes), scanned dimensions have been compared to measurements
Keywords:
performed with a calliper or a ruler: overall the values are mutually consistent. The proposed prototype is suit-
Linear infrastructure able for sewer inspections: displaced joints, cracks, deposits can be accurately measured without any bias by
Asset management comparison to CCTV. Uncertainty in the measurement appears to be unaffected by humidity or fat deposits.
CCTV 2016 Elsevier B.V. All rights reserved.
Camera
Laser
Uncertainty
Bias
Laboratory test
Validation
Prototype

1. Introduction
Sewers are aging underground structures that are often used beyond
their expected life duration and their actual functionality is hard to
quantify in situ. The management of such system requires accurate
and detailed data on their functioning and constructive strength and
stability in order to optimize the trade-off between the repair/replace-
ment costs and an acceptable level of various risks: urban ooding,
pipe collapse and/or pollution of water bodies caused by sewer over-
ows (environmental and public health risks). Various techniques al-
ready exist to inspect such linear infrastructure: i) CCTV inspection is
the most commonly applied in practise but has shown to be inaccurate
[2,3], ii) laser proling techniques [1,4], iii) acoustic techniques (e.g. in
[5,6]), or iv) a combination of them [7]. Previous studies highlight the
relative inaccuracy and the subjectivity of CCTV reports, due to: i) the
absence of measuring devices and/or software, ii) the quantication
classes of some standard methods [8,9], and iii) the human interpreta-
tion of the videos. CCTV inspections have been compared to acoustic
techniques [10] but not to laser scanning techniques: this is the main
goal of this study. Since laser inspection devices deliver 3D scans of a
Corresponding author.
E-mail address: m.j.lepot@tudelft.nl (M. Lepot).
sewer reach, they appear to be the most promising emerging technolo-
gy for accurate measurements [11]. Several industrial laser-scanning
devices are available on the market, based on two different approaches.
The rst one consists of a rotating laser distance meter, sometimes
mounted on a CCTV tractor (by example, the Rauschusa system [12]):
by construction, such a system just delivers a 3D image of the reach
only along a helical curve. The second approach, similar to the one
used in this study [1], is based on a ring laser light (a laser projected si-
multaneously at 360 degrees, e.g. the maverick Inspection system [13]).
In order to avoid bias in the data, both approaches require knowledge
on the exact position and spatial orientation of the device [4]: to the au-
thors' knowledge, no correction system is proposed by the commercial
systems so far.
The companion paper [1] describes in detail the design of the new
laser proler used in this study and provides important feedback and
suggestions for further developments: device accuracy, importance of
the laser alignment and a perfect synchronisation inter- and intra-ac-
quisition systems. Since the prototype has been built and calibrated, a
standard uncertainty of the air-solid interface position has been esti-
mated: from 1,1 mm to 1,8 mm, from invert level to the top of the
pipe, respectively. The prototype has to be exhaustively tested to ensure
its robustness and suitability for in situ measurements and applications.
Is the prototype able to measure fully the size of various objects in the

http://dx.doi.org/10.1016/j.autcon.2016.10.010
0926-5805/ 2016 Elsevier B.V. All rights reserved.
2 M. Lepot et al. / Automation in Construction 73 (2017) 111
pipe and to detect usual defects in sewer pipe? Under which conditions
does bias occur in the measurements? Does this technology present dis-
advantages when compared to CCTV inspections? In a more general
way, is the laser proler technique a potential candidate for replacing
CCTV?
In order to answer these questions and after a short introduction
into the presently applied inspection techniques and defect classica-
tion (Section 2), the design of the prototype is briey described
(Section 3.1), all the laboratory tests (performed on real pipes) are de-
scribed in Section 3.2. Data validation and analysis (comparison be-
tween measured - by a calliper or a ruler - and scanned - by the
prototype - object dimensions) methods are briey described in
Section 3.3. Results of the comparison between the scanned and mea-
sured sizes are presented and discussed in the fourth section of this
article.
2. Defects recorded by CCTV and their quantications
Sewer inspections are mainly done based on CCTV footage: defects
are characterised and quantied according to standard methods (vary-
ing however between EU, USA and other countries). As an example, a
European standard method [5] has been translated in national regula-
tions with some tiny differences per country. In Japan, defects identied
by CCTV are coded using a less detailed system of codes than in Europe.
The retained standard method for defect coding [8] records the fol-
lowing information (Table 1).
Based on the visual inspection and the position of the CCTV tractor,
the defects are positioned (longitudinal location - in m - from a refer-
ence point) and coded with different letters or numbers to:
dene the defect and its origin with the main code: deformation, s-
sure, break/collapse, defective brickwork or masonry, missing mor-
tar, surface damage, intruding connection, defective connection,
intruding sealing material, displaced joint, lining observation, defec-
tive repair, weld failure, porous pipe, soil visible through defect, void
visible through defect, roots, attached deposit, settled deposits, in-
gress of soil, other obstacles, inltration, exltration, vermin, other
connection (pipe), point repair, curvature of sewer, start node
type, nish node, general photograph, general remark, inspection
terminated before nish node, water level, ow from incoming
pipe, atmosphere within the pipeline and loss of vision;
characterise the defect, with specic sub-codes (1 and 2) depending
on the main code. For example, for a ssure (main code BAB), char-
acterisation 1 can be A for a surface crack, B for a crack or C for a frac-
ture and the characterisation 2 can be A for longitudinal, B for
circumferential, C for complex, D for helical or E for starting from a
point;
quantify the defect with a specic sub-codes (1 and 2) depending on
the main code. For example, a connection is quantied by Quanti-
cation 1 (in mm, the height of the connecting pipe) and Quantica-
tion 2 (in mm, the width only if it differs from the heights). Most of
the defects only have one quantication;
locate the beginning and the end of the defects according to the clock
face references (from 1 to 12);
indicate if the defect is located at a joint (transition) in the structure.
The previously listed defects highlight several advantages and disad-
vantages of the CCTV method. CCTV is only t for identifying defects that
can be visually detected by an operator. The quantication is done by
Table 1
Coding system of the European standard method of defect coding [5].
Longitudinal location Continuous defect code Main code Characterisation
1 2
human interpretation of images and, sometimes, with the position of
the tractor: subjectivity can lead to some bias in the data [2]. Measure-
ment by a laser proler seems to result in more objective results relating
more directly to the actual status of the sewer. However, some of the de-
fects can be detected only with colour camera and under good light con-
ditions such as lining observations, exltration or atmosphere-related
issues (i.e. presence of potentially hazardous atmosphere due to fog,
the occurrence oxygen deciency, hydrogen sulphide, methane or
other gasses).
3. Material and methods
3.1. The laser-proling prototype
The prototype consists of two main parts (Fig. 1): a stationary part
where three laser distance meters (Dimetix, FLS C10) and a pulsing
laser (Osela, Streamlaser 660 nm) have been set up and a moving
part where two cameras and a laser proler (Ibak, ILP) have been
installed.
The data stemming from the image processing of the front camera
(Allied Vision Technology, Prosilica GT3400C) frames (records of pro-
jection of the laser proler) are corrected with the exact position of
the moving part. The position is obtained from the three recorded dis-
tances and the image processing of the back camera (Allied Vision Tech-
nology, Manta G-282C) that records the projections of the four lasers set
up on the xed part. A detailed description on this design has been pre-
sented in [1]. The fourth laser has been used for an optical synchronisa-
tion between data acquisition system of both parts.
3.2. Object and material set
Based on the defects identied by [8], several add-ins have been set
up in an egg-shaped pipe (400/600 mm) to cover the wide range of de-
fects or solid materials, which can be found in sewer systems. Various
objects and materials (Table 2) have been used to study: i) the defects
detectable and quantiable by the prototype, and ii) effect of size,
shapes, colours, roughness or the kind of deposits on the measurement.
Pictures of the material set are available in the electronic appendix. Each
object has been scanned with the prototype: the moving part has been
manually moved into the pipe.
Water {1} and Vaseline {2} (Table 2) have been applied inside the
pipe, on the wall, in order to create conditions that may occur in sewers:
water saturated atmosphere (temperature, humidity, diffuse inltration
through the wall) and fat deposits. Other experiments have been done
to reproduce the most common sewer conditions: deposits of various
materials (as sludge {6}, sand {5}, gravels {7}, etc.), displaced joints
{3}, intrusive connections {12}, roots {13} and cracks {4}. Other objects,
more unusual but that can still be found in CCTV reports (like brick {16}
or a beer bottle {21}.), have been placed in the sewer pipe to test the
measuring capacities of the prototype.
3.3. Methods
3.3.1. Raw-data treatment
Previous studies [1,11,14] have presented the methods applied in
post-processing of the raw data (images and distances); these details
are not presented in this paper. Fig. 2 summarizes all the steps of the
data processing: from the raw data and sensors (left part) to the con-
struction of the 3D model by integration, from frame 1 to frame N, of
corrected (calibration), unbiased (correction of the position) and
Quantication Circumferential location Joint Photo ref. Video ref. Remarks
1 2 1 2
 M. Lepot et al. / Automation in Construction 73 (2017) 111 3

Fig. 1. Principle scheme of the prototype - left: moving part, right: xed part - top: side view, bottom: back view.

Table 2
Inventory of the objects and materials used, identied by a number X (noted {X} hereafter). For the experiments: name, material, colour and comments.

{Number} Group name Name Material Colour Description

{1} Wet Tap water Sprayed on the wall

{2} Vaseline Vaseline Applied manually on the wall
{3} Joint Weather stripping Black Non-constant thickness
{4} Cracks Same as the pipe Done with a drill and a sledge hammer
{5} Sediments Sand Sand Beige Placed on a wood board
{6} Sludge Earth + water Maroon Done with vegetal earth and tap water
{7} Sediments Gravels Rock Grey 58 mm
{8} Coloured stones Rock Blue, red and yellow Painted 1622 mm
{9} Sediments Small stones Rock Grey 2230 mm
{10} Sediments Medium stones Rock Grey 3040 mm
{11} Sediments Large stones Rock Grey 4063 mm
{12} Metal Various metal and cylindrical shapes Metal Various New and oxidized
{13} Roots Woods
{14} Small half pipe PVC Grey Cut part down, on {15}
{15} Large half pipe PVC Grey Cut part up
{16} Brick and concrete Brick Terracotta Brown
{17} Tiles Brown tile Terracotta Brown Bright
{18} Tiles White tile Terracotta White Bright
{19} Tiles Zebra tile Terracotta Beige & grey Bright
{20} Tiles Blue tile Terracotta Blue Bright
{21} Beer Glass Transparent green Beer bottle (33 cl)
{22} Metal Battery Steel Gold - blue LR20-D-SISE 1,5 V F401 NOVEL
{23} Metal Tap & valve Steel 1/2" 3mLBS Made in England
Oxidized
{24} Screw and nut Steel Grey Fuchs 4,6a
{25} Metal U bar Stainless steel Grey Thickness: 3 mm
{26} Brick and concrete Deltares tetrahedrons Painted concrete Red and yellow
{27} Brick and concrete Bollard Painted concrete Beige See Fig. 3
{28} Spanners Steel Grey Sizes 32 and 33 mm
{29} Lead Sheet Lead Grey Folded lead sheet
{30} Wheelbarrow inner tube Rubber Black Set up in the middle of the pipe
{31} Mixed Various Various Everything together
a
Model and size of the key.
4 M. Lepot et al. / Automation in Construction 73 (2017) 111
Fig. 2. Graphical summary of the applied methods for data treatment (LDM for Laser Distance Meter, d for distances, cameraBACK is the back camera, cameraFRONT is the front camera).
synchronised data. Relevant issues about the clock synchronisation
have been discussed in [1].
3.3.2. Object sizes: measurements and/or information available
The objects (listed in Table 1) have been measured with a calliper
(Gamma, Digital Calliper), which provide a standard uncertainty of
0,05 mm. Dimensions larger than 150 mm have been measured with a
simple ruler: the standard uncertainty for such measurement is estimat-
ed to be 0,5 mm. Reading errors (human observation uncertainty) have
been also taken into account: nally retained standard uncertainties of
measured distances are respectively estimated to be 0,1 mm and
1 mm for the calliper and the ruler.
Various materials have been used as natural solids: sands gravel
and stone (Table 1). Other shapes or articial solids, also used for scaled
hydraulic experiments, are shown in the electronics appendix: their
shapes and sizes are accurately known.
3.3.3. Comparison methods
For the CCTV and laser proling comparison, simulated defects
(Table 2) have been coded according to [8]. Based on pictures and labo-
ratory notes, the codes are presented in Table 3.
For the measurement comparison, the scanned (SizeSCANNED) and the
measured (SizeMEASURED) sizes, independently obtained, have been com-
pared with a simple variance test: if Eq.(1) is satised, the prototype
provides a measurement consistent with the one given by the calliper.
If not, the laser prototype provides biased estimations of object size.
The equation compares the variance (difference) between real (mea-
sured) and scanned sizes of the object and its uncertainty (at 95,4%):
the right part is derived from the law of propagation of uncertainties
[15] by [16].
q differentiable. The visible images furnished by CCTV seem to be more
jSizeSCANNED SizeMEASURED j 2  u2 SizeSCANNED u2 SizeMEASURED 1
where u2(SizeSCANNED) and respectively u2(SizeMEASURED) are the stan-
dard uncertainties associated to the laser and respectively the calliper
measurements. All variables are in mm. This equation has been applied
for all scanned objects.
The consistency ratios (RC) of each group (Table 1) have been calcu-
lated for all groups (Eq. (2)).
NC
RC 2
NT
where NC, is the number of objects or experiments of the group (com-
posed by NT objects of experiments) for which the Eq.(1) is satised.
All variables are dimensionless.
3.3.4. Roughness calculations
Different materials, grouped as sediments (Table 1), have been past-
ed on a PVC half-pipe (internal diameter of 280 mm), cut in half along
the main axe (half-circle prole), to simulate different sediment bed
compositions. This board has been scanned and roughness calculations
have been performed using the methods described in [7].
4. Results and discussion
4.1. Comparison between CCTV and laser proler
Table 3 presents a comparison between information that can be ex-
tracted from visual inspection (coded according to [8]) and laser
scanning.
For a wet wall {1} or the presence of fat deposits {2}, no disturbances
in the processes and the calculations have been observed in these ex-
periments: a moist and/or fat covered pipe can be scanned without dif-
culties. For a moist and/or fat pipe a small adjustment in the image
processing is required: since the recorded light intensities are smaller
the threshold on pixel value needs to be adapted. However, this adapta-
tion has no impact on the accuracy on the result. Diffuse inltration and
fat deposits are detectable due the light intensities, but are not mutually
suitable than laser scanning for identifying inltration and fat detection.
For some experiments, no quantication has been attempted. The sedi-
ments (objects from {6} to {11}) can only be quantied by the median
grain size the sizes of each stones or gravels (if presents) have not
been measured: only the overall median grain size. The sediments can
be qualied only using a standard camera and visual inspections: the
laser proler is unable to distinct the type of sediment. Some objects

Table 3
CCTV vs Laser scanning. CCTV results are expressed according to the following format: Main code//Characterisation 1/Characterisation 2//Quantication 1/Quantication 2//Circumferen-
tial location 1/Circumferential location 2//Joint//Remarks.
{Number} Description and CCTV
inspection code according to [8]
{1} Sweating inltration
BBF//A//0 to 12//-//
{2} Attached deposit (grease)
BBB//B//0//3//-//
{3} Displaced joint (radial)
BAJ//B//~10 mm//112//-//
{4} Complex fracture
BAB//C/C//~10 mm//9/2// //-//
Missing piece of wall (collapse)
BAC//B//~100 mm//3//-//
{5} Settled ne deposits (sand)
BBC//A//~3%//6//-//
{6} Settled ne deposits (sludge)
BBC//A//~8%//6//-//
{7} Settled coarse deposits (gravels)
BBC//B//~5%//6//-//
{8} Settled coarse deposits (rubble)
BBC//B//~5%//6//-//colours
{9} Settled coarse deposits (rubble)
BBC//B//~5%//6//-//
{10} Settled coarse deposits (rubble)
BBC//B//~10%//6//-//
{11} Settled coarse deposits (rubble)
BBC//B//~15%//6//-//
{12} Intruding connections (3)
3 BAG//~210%//6//-//
{13} Independent ne roots
BBA//B//~15%//1//-//
{14} Piece of drain lying on the invert
BBE//B//~8%//6//-//
{15} Piece of drain lying on the invert
BBE//B//~10%//6//-//
{16} Brick lying on the invert
BBE//A//~15%//6//-//
{17}, Other objects lying on the invert
{18}, BBE//C//~1.5%//6//-//
{19},
{20}
M. Lepot et al. / Automation in Construction 73 (2017) 111 5
Laser scanning data Comments
Not visible Wet wall (sweating inltration or moisture)
Effect on the light intensity Laser beam light intensity is weaker
Not visible Laser beam light intensity is weaker
Effect on the light intensity
Object detected, joint partially pending The joint can be identied with the pending part and its shadow.
Size (W = 22 mm, T = 211 mm) No material identication possible (need CCTV)
Location: 112
From 126 mm to 163 mm slightly tilted
Object detected, ssures and break The crack is clearly visible. Length, width and location can be accurately estimated.
Size - crack (width 411,5 mm; (big) crack The depth estimation is more sensitive to the relative position of the cracks and
depth 33 mm) the laser beam.
Location: from 2 to 9
Over the whole pipe length;
Size - break (61,5 102 mm)
Location 3
From 360 mm to 462 mm
Diagonal orientation
Object detected, deposit (ne material) Smooth surface (sand, sludge, fat deposit?)
Volume of 46,5 cm3 Quantication possible and consistent with the mass.
Location 6 Visible images (CCTV) are required to identify the sand.
From 53 to 355 mm
Object detected, deposit Coarser than {5}
Not attempted quantication
From 13 mm to 330 mm
Object detected, deposit No effect detected due to the colour of the stones
Not attempted quantication
From 28 mm to 290 mm
Object detected, deposit Coarser than {5}
Not attempted quantication; from 63 mm
to 366 mm
Object detected, deposit Coarser than {8} (Table 2)
Not attempted quantication from 53 mm
to 355 mm
Object detected, deposit Coarser than {9}
Not attempted quantication
From 22 mm to 279 mm
Object detected, deposit Coarser than {10}
Not attempted quantication
From 53 mm to 355 mm
3 objects detected, intruding connections Pipes easily identiable with the cylindrical shapes.
(shapes) Quantications accurate. Material non identiable without CCTV (Fig. 4, middle).
Sizes
(DIN = 54,6 mm, H = 99,7 mm; DEXT =
35 mm, H = 90 mm; DEXT = 54,1 mm,
H = 19,4 mm)
Location 6
From 35 to 89 mm
Object detected, roots or spider net The polygon dened by the roots is easily quantiable. Root sizes (diameter)
Size (area occupied: rectangle of 170 mm should be measurable (not tested)
width, 361,5 mm height)
Location from 1 to 3
From 1000 to 1221 mm
Object detected, most likely a pipe Pipe parallel to the sewer axis. Material not identiable without CCTV.
Size (H = 83,1 mm,
DEXT = 122 mm)
Location 6
Over the whole length
Object detected, most likely half-pipe Half pipe parallel to the sewer axis. Material not identiable without CCTV.
Size (H = 130 mm
DIN = 275,2 mm)
Location 6
Over the whole length
Object detected, rectangular cuboid (most Rectangular shape identiable. The clear identication requires CCTV image.
likely brick)
Size (W = 104,8 mm, H = 46,8 mm)
Location 6
From 141 mm to 223 mm
Objects detected, rectangular cuboids Same as {16}
Size (L = 97,3 mm, T = 11 mm) Tiles placed continuously, the transition is not visible.
Location 6
Starting at 45 mm, ending at 239 mm, from
239 to 343 mm, from 343 to 450 mm
(continued on next page)
6 M. Lepot et al. / Automation in Construction 73 (2017) 111

Table 3 (continued)

{Number} Description and CCTV Laser scanning data Comments

inspection code according to [8]

{21} BBE//C//~12%//6//-//beer bottle Something detected, crest or artefact The identication of the bottle is almost impossible. The data look like an artefact.
(bottle?) Identication impossible without CCTV.
Size (not clearly measurable)
Location 6 to 7
At 151 mm distance
{22} BBE//C//~2%//6//-//battery Object detected, cylindrical shape (battery?) Battery detectable but not identiable. Confusion possible with a piece of pipe
Size (DEXT = 30,7 mm, H = 61 mm) (small diameter).
Location 6
From 420 to 450,7 mm
{23} Other object lying on the invert Object detected The object is too complex to be identied with laser data.
BBE//C//~4%//6//-//tap and valve Not attempted quantication CCTV is clearly better for this kind of object.
Location 6
From 286 to 417 mm
{24} Other object lying on the invert Object detected Identication of the screw and the nuts is strongly dependent of their position.
BBE//C//~1%//6//-//screw and Location 6 CCTV is better for identication.
nut From 224 to 275 mm
{25} Other object lying on the invert Object detected, u-shaped object bar (one The shape is easily identiable. The material identication requires CCTV.
BBE//C//~10%//6//-//U bar plateau, two crests)
Size (H = 40,7 mm, W = 115,2 mm)
Location 6
Over the whole length
{26} Other object lying on the invert Object detected, tetrahedrons The shape can be identied. No effect of colours.
BBE//C//~10%//6//-//tetrahedron Size (Leg = 43,4 mm, DEXT = 19,3 mm)
Location 6
From 249 mm to 439 mm
{27} Other object lying on the invert Object detected, Bollard Due to the size and the position, the object is clearly identied. For less laser
BBE//C//~17%//6//-//bollard Size (Fig. 9) friendly position, CCTV might be required.
Location 6
From 105 to 225 mm
{28} Other object lying on the invert Object detected, most likely spanners Spanners are clearly identied.
BBE//C//~4%//6//-//spanners Size (U-opening 3233 mm)
Location 6
From 50 to 461 mm
{29} Other object lying on the invert Object detected, deposits No distinction can be made between deposits and this lead sheet CCTV is required
BBE//C//~15%//6//lead sheet Not attempted quantication for identication.
Location 6
Starting at 93 mm
{30} Other object lying on the invert Object detected, an inner tube (shape, Section reduction (at 360 degrees or less) is detectable and accurately
BBE//C//~60%//6//-// inner valve) quantiable.
tube + piece of wood Size (DIN = 189,7 mm) End point not clearly measurable because the shadow effect
Location: the middle of the pipe Reduction of 86,5%
Starting at 123 mm

({23}, {29}) were too complex to provide quantication: the data ob-
tained were not good enough to deliver a meaningful measurement re-
sult (especially for the tap {23}).
Displaced joints and section reductions have been simulated with
experiments {3} and {30} (Fig. 3).
Fig. 3 clearly shows that these defect are detected by the prototype:
the proler has no difculties to scan both objects. The overall shapes
seem to be correctly reproduced in the data (including the valve of
the inner tube). However, shadow effects (blind zones) are visible (an
arc and a ring behind the displaced joint, a ring around the inner
tube). For the displaced joint (Fig. 3, bottom left), the pending part hin-
ders the measurement of the top part of the pipe after the joint (Fig. 3,
middle left). The camera cannot see behind this pending part. The
same issue persisted with the back part of the inner tube. Shadow ef-
fects may be also attributed to a gap in the laser projection. This is clear-
ly visible with the inner tube: the wood piece, placed to maintain the
inner tube in its position, is not visible: the inner tube has blocked the
front camera's vision of the laser projection. The wooden part is indi-
rectly visible: the additional gap right after the valve. Despite this weak-
ness, which is not present in the latest version of CCTV tractor equipped
with two cameras, the quantication and the location of the defect can
be accurately determined (Table 3). CCTV inspections still present one
extra advantage: the pending object (joint) and the object responsible
of the section reduction are clearly identiable with CCTV images/
videos. Those experiments have highlighted some major issues: i) to
limit or avoid shadow effects, both camera and laser proler sources
should be multiple and ii) potential information can be extracted from
gaps in the data. These outcomes may initiate developments of new
methods for pattern recognitions.
Intrusive connections have been simulated with objects {12}, {14}
and {15}. The prototype has been able to detect, locate and quantify the
three connections (Fig. 4 and Table 3). Laser scanning is more accurate
than CCTV measurement is this respect. Furthermore, the prototype has
provided consistent measurements (Table 4) for the diameter of the var-
ious cylindrical shapes. Used in real sewers, laser scanning will be able to
locate and quantify accurately intrusive connections. The kind of connec-
tion (depending on the diameter) may be determined. Full (1st and 2nd
pipes) or empty (third one) can be measured and distinguished (circles
or disks visible in the scanned data): the prototype is able to detect if
there is (are) solid(s) or blockage(s) in the connected pipe.
Roots have been simulated with the object {13}: ne roots have been
glued to pipe crown. Fig. 5 presents the scanned data.
Roots and their main directions are visible in the scanned data. The
contour or the area occupied by such roots can be easily assessed
(Table 3). However, visible images from CCTV will provide more details
on the roots and their entry points, missing a proper quantication
though.
An object's colour does not affect the measurement. Only transpar-
ent (glass bottle {21}) or highly reective (tile as {20}) objects disturb
the measurements as shown in Fig. 6 (glass bottle {21}).
The device has only been able to scan the top crest of the bottle:
transparent objects create diffraction and distorted shadows while
 M. Lepot et al. / Automation in Construction 73 (2017) 111 7

Fig. 3. Displaced joint {3} (left) and section reduction with one inner tube {30} (right), picture (top), camFRONT screen shots (post-processed with Photoshop, middle) and scanned data
(bottom). Blind zones are clearly visible (middle left): real picture, without Photoshop treatment are visible in the electronic appendix - high contrast).

the projected laser light is passing through the object, only the per-
pendicular parts (to the projected laser) of transparent objects may
be measurable with such prototype. The distorted shadow leads to
gaps in the 3D data (Fig. 6). As for bright objects, transparent ones
are relatively uncommon in sewers. This is a clear disadvantage of
laser proling when compared to CCTV inspection footage.
The laser proling technique allows for quantication of the volume
of deposits, this is achieved - in the laboratory - by scanning the pipe
twice: once with sediments and a second time after the sediments has
been removed. For in situ application, the part of the pipe covered by
sediment needs to be assumed i.e. supposed equal to the theoretical sec-
tion: this will lead to additional uncertainty and maybe, to some non-
8 M. Lepot et al. / Automation in Construction 73 (2017) 111

Fig. 4. Various cylindrical shapes scanned with the device: photo (left), screen shot of the front camera video (post-processed with Photoshop, middle) and scanned data (right).

discovered defect hidden by the sediments. The experiment {5} (with
and without sand) has been carried out (Fig. 7). First, the pipe with
the wooden board has been scanned. After adding about 67 g of sand
(approximately 46,5 cm3 based on typical loose sand density of
1442 kg/m3) on the board (Fig. 7, left), the pipe has been scanned
again. The difference between the two results is due to the deposit
and has been plotted in Fig. 7 (right). The difference between consecu-
tive measurements without and with sand present gave the following
estimation of the volume V = 45,2 cm3. The deviation of the measured
value from the actual is in the order of 3,5%.
Sludge {6} deposits have also been scanned by the prototype and did
not lead to difculties in the data treatment: this outcome is consistent
with the ones previously established (fat and wet wall). As with sand
deposit, the sludge level can be measured. The laser scanning technique
delivers better results when compared to CCTV inspection in terms of
quantifying levels and volumes: a 3D map of the deposit can be com-
posed allowing for the deposit volume to be calculated.
Other defects codied by [8] have been previously tested in [11].
Section deformations (coded as BAA) can be seen as erosion and deposit
on some clock faces. For example, vertical deformation of a pipe (BAA//
A) will be identied by a lack of material at the invert and the crown re-
gions and some added material at clock face references 3 and 9. As dem-
onstrated by [4], such a prototype will be able to detect and quantify
pipe deformation. CCTV inspection still presents a major advantage:
identication of the likely cause of the deformation (material erosion,
sediment deposit or a deformation). For the same reason, surface dam-
age (coded BAF according to [8]) can be detected, quantied but not
characterised by laser scanning techniques. The pipe alignment can be
easily checked and misalignments quantied by construction a continu-
ous 3D gure of reach composed by several pipes (Fig. 8, from [1]). The
white spaces are missing data: shadows effects explain the constant
thickness ones (rst space on both gures), pitch, roll and/or yaw angles
from the proler explain the not triangular ones (e.g. from z = 3900 to
4200 mm in the bottom gure). Lack of contrast (at the junction - inlet
of light) also explains some blanks in the Fig. 8. High y values (bottom
graph) are due to a hole in the crown region and transition between
pipes (space between male - female plug).
There are still some defects, coded by [8], which cannot be detected
convincingly while even characterisation or quantication by the laser
based technique poses a problem: i.e. lining observation, soil or void vis-
ible through the defect and exltration. These defects and some limita-
tions previously established lead to the conclusions that cameras
providing coloured pictures are still mandatory to get a good estimation
of the current pipe status. Laser proling cannot be regarded to be self-
sufcient for sewer inspection.

Table 4
Results of the comparison tests.

{Number} Compared Consistent Not-consistent

dimensions dimensions dimensions

{3} W, T W, T
{12} DIN, H DIN, H
DEXT, H H DEXT
DEXT, H H DEXT
{15} DIN DIN
{16} L, W, H L W, H
{17} L, T L, T
{18} L, T L, T
{19} L, T L, T
{20} L, T L, T
{22} DEXT, H, L H DEXT, L
{25} H, L H L
{26} Leg, DEXT Leg, DEXT
{27} See Fig. 9
{28} DEXT, DEXT DEXT, DEXT
Fig. 5. Roots scanned by the prototype.
 M. Lepot et al. / Automation in Construction 73 (2017) 111 9

Fig. 8. Top (up) and side (down) view of the results of 3D reconstruction of the scanned
sewer pipe stretch (from [1]).

values given in Appendix A highlight that the maximal difference be-

Fig. 6. Scanned data of the glass bottle.
4.2. Robustness of the laser proler measurements
The following results highlight the robustness of the measurement
done by the prototype in order to conrm the representativity of vol-
ume deposit measurement. The results of the scanned/measured size
comparison are presented in Table 4. Measured and scanned dimen-
sions (L: length, W: width, H: height, T: thickness, DEXT: external diam-
eter, DIN: internal diameter, Leg: length of the leg of the tetrahedron, To:
the length of the tooth and B the space between two teeth) are present-
ed in detail in Appendix A. Compared dimensions are summarized
hereafter and results of the test (Eq. (1)) are given (C: consistent, I:
inconsistent).
Table 4 shows that most of the dimensions are consistent: the laser
proler provides robust estimations of the main object sizes. Detailed
tween scanned and measured values is about 7 mm.
The consistency ratios of the groups as dened in Table 2 are pre-
sented in Table 5. For concrete objects, the consistency ratio is 047
and, if the distance along the movement direction is not considered,
this ratio increases up to 0.67. Metal objects lead to the same conclu-
sions, with higher consistency ratios. For the tiles, the consistency
ratio is 1. For those objects, every dimension (except the one along
the main axis of the pipe) can be accurately measured: the results are
consistent with dimensions obtained by using a calliper.
These ratios emphasize the relatively low resolution obtained along
the z-axis. Due to the construction and the way to use such a prototype,
the data density (i.e. frame resolution) along the z-axis is directly

16
Sand deposit
250 (mm)

14

12
200

10

150
8
z (mm)

6
100

2
50

20 40
x (mm)

Fig. 7. Quantication of the sand deposit volume: photo of the experiment (left) and Fig. 9. Scanned (normal font) and measured (italic) sizes of the bollard. Consistency (C) or
scanned sand relief (right). Inconsistency (I) between the two sizes.
10 M. Lepot et al. / Automation in Construction 73 (2017) 111
Table 5
Consistency ratios of the groups (only directions perpendicular to
the main pipe axis).
Group names RC
()
Bricks and concrete 0,47 (0,67)
Tiles 1
Metal 0,67 (0,875)
dependent on two parameters: the frame rate of the front camera
(12 fps for the presented prototype) and the speed of the moving part
(23 cm/s). The desired speed was 4 cm/s that leads theoretically to a
frame every 3,3 mm (1,72,5 mm in reality): this value is not constant
and varies from frame to frame due to non-constant speed, pitch, yaw
and roll angles. The low resolution along the z-axis may be responsible
for some inconsistencies (see experiments {12}, {22}, {27} or {28}) be-
tween scanned and measured sizes. In order to reduce this weakness
of the presented prototype, two solutions can be applied: i) increase
the image acquisition frequency (to e.g. 24 or 36 fps) and/or ii) reduce
the speed of the moving part. The rst solution will increase the prob-
lem encountered with the data ow as highlighted in [1]. The second
one is feasible for research purposes, not for routine applications. This
discussion emphasizes the need of image pre-processing in order to re-
duce the required data transmission and storage capacity while improv-
ing accuracy of the results along the main axis of pipe.
4.3. Roughness measurements
For deposits built with known materials ({5}, {7}, {9}, {10} and {11}
of Table 2), roughness calculations (as described in [14]) have been per-
formed and their results are summarized in Table 6.
Due to the lack of accuracy and the low resolution, standard uncer-
tainties (u(d50)) derived from the laser proler accuracy and interpola-
tion errors [14]) are of the same order of magnitude (mm) as the
median grain size (d50) itself: the prototype is not sufciently accurate
to estimate roughness. A third version of this laser scanning technology
is required for in situ roughness measurements. By applying the same
calculations as in [1] and to obtain a target accuracy of 0,15 mm, the fol-
lowing theoretical improvements should be implemented: laser dis-
tance meters should be better aligned (up to 1000 times more
accurate), distances provided by these devices should be 30 times
more accurate, camera accuracy (see [1]) needs to be improved by a fac-
tor of 10, the camera offset and scaling factor should be 100 times more
accurate. Relative contributions to the total uncertainty have been plot-
ted in Fig. 10. The theoretical third version seems to be infeasible in
2016: in situ roughness measurements with laser proling devices are
unlikely to achieve this accuracy on the short term; at the current status
it is possible to use laser based techniques to quantify wall roughness
using high precision 2-D laser scanning techniques as described in [14].
5. Conclusions and perspectives
After the work done by [1] (design and assessment of the standard
uncertainties), the second part of the paper has summarized various ex-
periments. This study underlines relevant outcomes for further research
Table 6
Results of the roughness calculations.
Material Granulometry d50
mm
Sand 0,1 mm 1,74
Gravel 58 mm 6,13
Small stones 2230 mm 11,25
Medium stones 3040 mm 15,62
Large stones 4063 mm 17,34
Fig. 10. Expected accuracy of a theoretical third version of a laser-scanning device to reach
a standard uncertainty of the position of the cross-section of 0,15 mm. Legend: bx,y,1 x, y
coordinates of a point as observed on the optical sensor of camFRONT, bx,y,2 x, y
coordinates of a point as observed on the optical sensor of camBACK, dz1 distance
between laser sheet and lens plane for camFRONT, dz2 distance between laser sheet and
lens plane for camBACK, ez1 distance between optical sensor and lens plane for
camFRONT, ez2 distance between optical sensor and lens plane for camBACK.
on this topic. First, most of the scanned sizes are consistent with the
measured ones. Differences are always lower than 7 mm and can be par-
tially explained by the data or frame density along the main axis of the
pipe: increasing the image acquisition frequency or decreasing the
speed may reduce this weakness. No bias has been identied: the meth-
od seems to be insensitive to colours or materials. Only reective or
transparent objects create disturbances in the measurements. Fortu-
nately, such objects are not very common in sewer systems. Unbiased
proong techniques, associated with an intensive data treatment, can
provide accurate and robust information. Most of the common defects
can be quantied: deposits, wall erosion [4], displaced joints, section re-
duction, intrusive connections and roots. The reliability of this tech-
nique has been demonstrated.
Some weaknesses, when compared to CCTV inspection, need to be
highlighted. Firstly, there is no real-time data transmission: an operator
cannot adjust velocity and behaviour of the moving part. Real-time data
transmission (wireless) does not seem easy to implement regarding the
data ow generated by the cameras applied. Basic defect detection algo-
rithms (as [17,18,19,20]) may be implemented in a future version in
order to adjust the moving speed next to a defect. Secondly, there is
no light on the prototype and due to the high contrast in the eld of
view (laser in darkness, Fig. 4 in the middle): colours are not visible
and an operator cannot qualitatively estimate the defects. In order to
compensate this weakness, a low-resolution camera and LED lights
could be mounted on a future version of the prototype. An automatic
defect detection algorithm can control this camera. This optical tech-
nique is also still sensitive to shadow: multiple laser prolers and
front cameras may theoretically reduce or solve this problem. In prac-
tise, it will not be easy to implement, especially due to the amount of
data which will become even more difcult to manage. Subsequent
studies could be performed: i) a better end-user data treatment (in
order to quantify section reduction (in m2), length of pending joint (in
m)), ii) a more accurate device in order to reduce the uncertainty
along the z-axis and/or to measure in situ the wall roughness and iii)
some pattern recognition algorithms based on both existing data and
gaps to automatically recognize and quantify the most common defects.
Finally, the tested prototype was a research version, not an end-user
u(d50) version: some additional work will be required to build a waterproof
mm and solid version (as structure to support hardware and the connections
2,01 between apparatus).
2,26 At present this technology is not really mature enough to fully re-
2,68 place CCTV but offers some advantages over visual, CCTV based inspec-
3,07 tion: accuracy, repeatability and objectivity. CCTV inspections have a
3,66
serious disadvantage: the qualication on the defect based on human
 M. Lepot et al. / Automation in Construction 73 (2017) 111 11

interpretation of images. Further improvements are still required to Table A.1 (continued)
make this method a new standard instead of CCTV. However, a device {Number} Measured Scanned Difference Extended uncertainty Result
for accurate in situ roughness measurements appears to be unrealistic dimensions dimensions (mm) (right part of Eq. (1)) (C or I)
at present using the presented laser scanning technology: calculations (mm) (mm) (mm)
have shown that to this end very accurate devices (camera and laser dis- {27} See Fig. 9
tance meters) and construction processes (to align the lasers and know {28} DEXT: 31.7 DEXT: 32 DEXT: -0,3 DEXT: 3,2 DEXT:
accurately the camera offset) are required to reach a standard uncer- DEXT 33,3 DEXT: 33 DEXT: 0,3 DEXT: 3,2 C
DEXT:
tainty of 0,15 mm. Other solutions can be studied in order to increase
C
the accuracy: high quality camera-lenses, large camera sensors, thinner
a
laser beams or a rotating 2D laser scanning (more expensive). n.c.: not calculated.

Appendix B. Supplementary data

Acknowledgement

Supplementary data to this article can be found online at doi:10.

The authors would like to acknowledge the funding by (in alphabet-
1016/j.autcon.2016.10.010.
ical order) ARCADIS, Deltares, Royal Haskoning DHV, Municipalities of
Almere, Breda, 's-Gravenhage, Rotterdam and Utrecht, GMB
References
Rioleringstechnieken, Grontmij, KWR Watercycle Research Institute,
Stichting RIONED, STOWA, Tauw, Vandervalk & De Groot, Waterboard [1] N. Stani, M. Lepot, M. Catieau, J.G. Langeveld, F.H.L.R. Clemens, A Technology for
De Dommel, Waternet and Witteveen + Bos as participants of the Sewer Pipe Inspection (Part I): Design, Calibrations, Corrections and Potential Appli-
cation of a Laser Proler(Submitted to Journal of Automation in Construction) 2016.
Urban Drainage Research program. In addition, Foundation RIONED is [2] J. Dirksen, F.H.L.R. Clemens, H. Korving, F. Cherqui, P. Le Gauffre, T. Ertl, et al., The
acknowledged for additional funds for purchasing the hardware needed consistency of visual sewer inspection data, Struct. Infrastruct. Eng. 9 (3) (2011)
for the prototype as presented in this article, Deltares is acknowledged 214228, http://dx.doi.org/10.1080/15732479.2010.541265.
[3] V. Sousa, F.M. Ferreira, I. Meireles, N. Almeida, J.S. Matos, Comparison of the inspec-
for making avaiable their laboratory to conduct the experimental work.
tor and rating protocol uncertainty inuence in the condition rating of sewers,
Water Sci. Technol. 69 (4) (2014) 862867, http://dx.doi.org/10.2166/wst.2013.
Appendix A 764.
[4] R. Henry, A.R. Luxmoore, A pipe-proling adapter for CCTV inspection cameras: de-
velopment of a pipe-proling instrument, Meas. Sci. Technol. 7 (4) (1996) http://dx.
doi.org/10.1088/0957-0233/7/4/005.
Table A.1 [5] A. Romova, Acoustic Monitoring of Hydraulic Resistance in Partially Full Pipes(PhD
Results of the comparison tests. thesis) University of Bradford, 2014 (240 p).
[6] M.T. Ali, K.V. Horoshenkov, S.J. Tait, Rapid detection of sewer defects and blockages
{Number} Measured Scanned Difference Extended uncertainty Result using acoustic-based instrumentation, Water Sci. Technol. 64 (8) (2011)
dimensions dimensions (mm) (right part of Eq. (1)) (C or I) 17001707, http://dx.doi.org/10.2166/wst.2011.183.
(mm) (mm) (mm) [7] H. Plihal, F. Kretschmer, M.T. Bin Ali, C.H. See, A. Romanova, K.V. Horoshenkov, T.
Ertl, A novel method for rapid inspection of sewer networks: combining acoustic
{3} L: 1400 L: n.c.a L: n.c. L: n.c L:
and optical means, Urban Water J. 13 (1) (2016) 314, http://dx.doi.org/10.1080/
W: 20 W: 22 W: 2 W: 3,2 W: C
1573062X.2015.1076857.
T: 310 T: 211,1 T: 1 T: 3,8 T: C [8] CEN (European Committee for Standardization), Investigation and Assessment of
1,1 Drain and Sewer Systems Outside Buildings Part 2: Visual Inspection Coding
{12} DIN: 52 DIN: 54,6 DIN: DIN: 3,2 DIN: C System(EN 13508-2+A1, Brussels, Belgium) 2011.
H: 100 H: 99,7 2,6 H: 3,2 H: C [9] NEN (Nederlands Normalisatie-instituut), Onderzoek en beoordeling van de
DEXT: 40 DEXT: 35 H: 0,3 DEXT: 3,2 DEXT: I buitenriolering - Deel 2: Coderingssysteem bij visuele inspectie (Investigation and
H: 91 H: 90 DEXT: 5 H: 3,2 H: C Assessment of Drain and Sewer Systems Outside Buildings Part 2: Visual Inspec-
DEXT: 50 DEXT: 54,1 H: 1 DEXT: 3,2 DEXT: I tion Coding System), 2004 (NEN-EN 13508-2+A1, Delft, The Netherlands, (2011)).
H: 19 H: 19,4 DEXT: H: 3,2 H: C [10] A. Romanova, K.V. Horoshenkov, S.J. Tait, T. Ertl, Sewer inspection and comparison of
4,1 acoustic and CCTV methods, Proc. Inst. Civ. Eng. Water Manage. 166 (2) (2013)
H: 0,4 7080.
{14} DEXT: 125 DEXT: n.c. DEXT: n.c. DEXT: n.c DEXT: - [11] F.H.L.R. Clemens, N. Stani, W. van der Schoot, J. Langeveld, M. Lepot, Uncertainties
associated with laser proling of concrete sewer pipes for the quantication of the
DIN: 120 DIN: n.c. DIN: n.c. DIN: n.c DIN: -
interior geometry, Struct. Infrastruct. Eng. (2014) http://dx.doi.org/10.1080/
{15} DEXT: 1060 DEXT: n.c. DEXT: n.c. DEXT: n.c DEXT:
15732479.2014.945466.
DIN: 1050 DIN: 1046 DIN: 4 DIN: 3,2
[12] https://rauschusa.com/products/laserprole (consulted the 3rd of November 2016).
DIN: I [13] http://www.maverickinspection.com/services/remote-video-inspection/laser-pipe-
{16} L: 209,8 L: 208,4 L: 1,4 L: 3,8 L: C proling/laser-pipe-proling-intro/ (consulted the 3rd of November 2016).
W: 98,3 W: 104,8 W: 6,5 W: 3,2 W: I [14] N. Stani, F.H.L.R. Clemens, J. Langeveld, Estimation of Hydraulic Roughness of Con-
H: 51 H: 46,8 H: 4,2 H: 3,2 H: I crete Sewer Pipes by Laser Scanning(Submitted to Journal of Hydraulic Engineering)
{17} L: 99,4 L: 97,3 L: 2,1 L: 3,2 L: C 2016.
W: 99,5 W: W: W: W: [15] JCGM 104, ISO/IEC Guide 98-1 - Uncertainty of MeasurementPart 1: Introduction
T: 8,3 T: 11 T: 2,7 T: 3,2 T: C to Expression of Uncertainty in Measurement, ISO - International Organization for
{18} L: 97,8 L: 97,3 L: 0,5 L: 3,2 L: C Standardization, Geneva (Switzerland), 2009.
W: 97,1 W: W: W: W: [16] GRAIE, Fiche mthodologique n2: Comparaison de deux valeurs, GRAIE, groupe de
T: 8,3 T: 11 T: 2,7 T: 3,2 T: C travail Autosurveillance des rseaux d'assainissement, Villeurbanne (France),
2008.
{19} L: 98,9 L: 97,3 L: 1,6 L: 3,2 L: C
[17] H.M.R. Halfawy, J. Hengmeechai, Integrated vision-based system for automated de-
W: 99,1 W: W: W: W:
fect detection in sewer closed circuit television inspection videos, J. Comput. Civ.
T: 9,3 T: 11 T: 1,7 T: 3,2 T: C
Eng. 29 (1) (2014) http://dx.doi.org/10.1061/(ASCE)CP.1943-5487.0000312.
{20} L: 97,0 L: 97,3 L: 0,3 L: 3,2 L: C [18] H.M.R. Halfawy, J. Hengmeechai, Efcient algorithm for crack detection in sewer im-
W: 97,2 W: W: W: W: ages from closed-circuit television inspections, J. Infrastruct. Syst. 20 (2) (2013)
T: 7,9 T: 11 T: 3,1 T: 3,2 T: C http://dx.doi.org/10.1061/(ASCE)IS.1943-555X.0000161.
{22} DEXT: 34,2 DEXT: 30,7 DEXT: DEXT: 3,2 DEXT: I [19] H.M.R. Halfawy, J. Hengmeechai, Automated defect detection in sewer closed circuit
H: 61,5 H: 61 3,7 H: 3,2 H: C television images using histograms of oriented gradients and support vector ma-
H: 0,5 chine, Autom. Constr. 38 (2014) 113, http://dx.doi.org/10.1061/(ASCE)IS.1943-
{25} L: 120 L: 115,2 L: 4,8 L: 3,2 L: I 555X.0000161.
H: 42 H: 40,7 H: 1,3 H: 3,2 H: C [20] O. Duran, K. Althoefer, L.D. Seneviratne, Automated pipe defect detection and cate-
{26} Leg: 42,0 Leg: 43,3 Leg: -1,3 Leg: 3,2 Leg: C gorization using camera/laser-based proler and articial neural network, IEEE
DEXT: 20 DEXT: 19,3 DEXT: 0,7 DEXT: 3,2 DEXT: Trans. Autom. Sci. Eng. 4 (1) (2007) 118126, http://dx.doi.org/10.1109/TASE.
C 2006.873225.