Robust Tracking of Circular Features

This paper considers the problem of internally inspecting underground pipes, such as sewers, in order to identify and locate major structural defects or damage to the pipes, and to identify potential defects which may lead to failure. It considers three aspects of the pipe inspection task; namely, the extraction of geometric primitives from the pipe image sequences, the tracking of these primitives over time, and the detection of (potential) structural collapse of the pipes. Knowledge of the environment (i.e. the reflectivity and geometry characteristics of the pipes) is exploited to develop appropriate methods for extracting the pipe joint features from edge information and to fit a pipe joint model (a circle, or set of connected arc segments) to the grouped features; and to track these features in sequential images. The paper describes a novel implementation of a Hough-based circle detection algorithm and compares its operation with a least-squares curve fitting algorithm for the detection of circular image features. Results of robust tracking of these pipe joints over a large sequence of image frames (some 1500 frames) are presented. Finally, these circle data are employed to aid the detection of partial pipe collapse.