A system for the practical use of computer vision techniques is presen
ted for the automatic assessment of the structural condition of sewers
from video images. The system can provide accuracy, efficiency and ec
onomy of sewer pipe examination. The main efforts of the paper have be
en placed on investigations of algorithms and techniques for image pro
cessing, pipe-joint shape analysis and discrimination, and crack detec
tion. Experiments show that over 85% of the pipe-joint boundaries can
be extracted successfully from video images, and for pipe-joints with
very low aspect ratio (AP<0.05), the measurement error in AP is no mor
e than 0.01. This kind of accuracy is of great significance in sewer p
ipe structural surveys, and cannot be achieved by visual inspection of
the video images. (C) 1997 Pattern Recognition Society. Published by
Elsevier Science Ltd.