Proceedings
of
the
2003
IEEE
International
Conference
0.
Robotics &Automation
Taipei, Taiwan, September
14.19,
2003
Experiments using a Laser-based Transducer and Automated Analysis Techniques
for Pipe Inspection
Olga Duran, Kaspar Althoefer
and
Lahal
D.
Seneviratne
email
:
{olga.duran,
k,
althoefer,
la~al.seneviratne/@kcl.ac.
uk
Depamnent of Mechanical Engineering
King's College London
Strand, London WCZR
2LS,
United Kingdom
Absmcr
-
This paper presents the experimental results
of
an
automated sensor system fur the inspection of tubular struc-
tures. The method is applied to the autonomous inspection of
Sewers overcoming the drawbacks
of
standard CCTV-based
inspection systems. The transducer consists of
B
low-cost
laser-
based profier attached to
B
standard CCTV camera. Image
analysis techniques and artificial neural networks are used to
automatically locate and classify the defects in the pipe using the
intensity distribution in the acquired camera images.
A wide
range of tests using data from different types of pipes in realistic
conditions have been conducted and are presented here. It is
shown that the proposed inspection approach is particularly
well suited tu complement existing CCTV inspection systems,
providing automated and reliable detection
of
pipe defects in the
millimeter range.
I.
INTRODUCTION
Standard pipe inspection systems are based
on
Closed
Circuit Television (CCTV) cameras in a large range of appli-
cation fields such
as waste pipes and drains. The CCTV
method consists of a mobile, remotely-operated platform usu-
ally equipped with a color, high-resolution video camera and
a lighting system. The camera platform is connected via a
multi-core cable to a remote inspection station with video
recording facilities situated overground. An engineer then
assesses
the
recorded
images
off-line. This is a subjective and
time-consuming task that considerably increases the inspec-
tion costs. Moreover, only
gross anomalies are evident to the
human eye, which reduces the detection of faults at early
stages. Another drawback associated with those systems in
these particular environments is the lack of visibility inside
the pipes and the poor quality of the acquired images that
hinders a complete assessment
of
the pipe condition and
sometimes even the detection of large defects.
A number of automated inspection techniques that aim to
cope with the drawbacks of CCTV have been proposed in
recent years
[l].
Special lighting and profilers systems have
been proposed to cope with the image quality problems [1,4].
These systems usually work by projecting either a thin ring of
light
or
successive light spots using
a
rotating mechanism
onto the pipe wall.
T.
Tsuhouchi and
et
a1
proposed a differ-
ent approach using a laser spot array instead of a light ring
[ll].
As the platForm moves through the pipe, the succession
of profile measurements allows the creation of
a
surface map
of
the
inner pipe wall. Geometrical changes
of
the pipe
sur-
face can be retrieved from the changes in the position of the
ring on the acquired image using the principle of triangula-
tion [l].
Besides image quality problems, the automation
of
the
process has been proved to be
a
very important issue, and
intelligent classification defect algorithms are investigated
[l].
Recognition and classification of pipe surface defects
from digitized video images using image analysis pattern
recognition and artificial neural networks have been proposed
recently [2,3]. Although those systems overcome automation-
related disadvantages
of
human CCTV assessment, they still
rely
on
the quality of the raw camera images.
In this paper a laser profiler is used to enhance the quality of
the pipe images. Compared to previous work, the novelty
here
is
the use of the intensity information instead
of
the
positional deformation
of
the ring of light. Intelligent analysis
and classification techniques are applied to the acquired im-
ages to automate the inspection process. This approach is
based on detecting sharp changes in the image intensity val-
ues. Neural networks are used in
a
first stage for the identifi-
cation of defective pipe sections, and Mer to classify them
into types of defects. The input to the ANN
is
a pre-processed
signal that emphasizes the differences between defective and
non-defective pipe sections. The network employed
is
a
multi-layer perccptron (MLP), trained by a backpropagation
algorithm, widely used for solving classification problems
1731.
This paper extends the research that aims at creating
autonomous
sensors
for low-cost, self-reliant pipe inspection
presented in
[81.
Here a wide range of tests using data from
different types of pipes are presented. It is shown that the
proposed inspection approach is particularly well suited to
complement existing CCTV inspection systems, providing
automated detection and classification of pipe defects in the
millimeter range employing a low-cost system add-on and
classification algorithms with real-time capability.
0-7803-7736-2/03/$17.00 02003
IEEE
2561