Automated Analysis of Sewer Inspection Closed Circuit Television Videos Using Image Processing Techniques
Date
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Closed circuit television (CCTV) has been the primary sewer inspection method over
four decades, and it is still widely used around the world. Sewer condition information is
an important tool for developing asset management and renewal planning strategies.
Several other sewer inspection technologies, such as digital side scanners, sonar, and
laser-based scanning have rapidly advanced over the last two decades. However, these
technologies still have limited use in the industry, and CCTV technology remains the
most widely used sewer inspection technology.
CCTV sewer inspections are highly dependent on the operator’s interpretation and
assessment, and hence, could be somewhat subjective. Also, operators may be subject to
fatigue due to lengthy inspection sessions, which could lead to erroneous assessment of
the sewer condition. This research aims to develop algorithms and a software prototype to
automate the analysis and assessment of sewer condition from CCTV videos using image
processing and pattern recognition techniques. Several new algorithms are proposed to
support automated identification of regions of interest (ROI) in the CCTV videos,
classification of frames based on camera orientation, segmentation using grey-level
intensity analysis, and automated detection of several sewer defects.
Starting from a raw CCTV video, the system’s operation starts by performing camera
motion analysis to identify and locate the ROI inside the sewer. These ROI represent
suspicious
video segments where sewer defects are more likely to be present. Frames within the ROI are processed and analyzed to extract useful information and identify
existing defects, if any. A number of algorithms were developed to segment individual
frames and to automatically detect and classify structural and operational defects.
The system is composed of three main components. The first component was
implemented in C++ using Intel’s Open Computer Vision library. This component
provides users with a graphical interface that enables access to all functions provided by
other software components. The second component comprises a set of MATLAB scripts
that implement the segmentation and defect detection algorithms. The third component
includes the SVMlight software. The system was built and tested using a set of CCTV
videos obtained from the City of Regina, Canada