Feasibility of active vision for inspection of continuous concrete pipes.

This thesis describes work to establish the feasibility of using active vision on a mobile robot to improve survey techniques for concrete and clay sewers of less than 1m diameter. Software and hardware components of a prototype mobile remote visual sensing system have been designed and developed. The active vision system (AVS) operates within smooth-walled small-bore pipes (0.5m < d < 1.0m). The AVS consists of two distinct, but related hardware components, a controllable (pan and tilt) camera head mounted on a remote control tractor and a system control unit which interfaces this remote system to a PC-based system supporting image capture and analysis.The software associated with the AVS comprises modules to control the camera orientation and supplement existing Artificial Intelligence vision analysis tools. The latter modules estimate the vanishing point (VP) of a sewer pipe (as a reference feature) and detect coaxial cracks in the periphery of the image (nearest the camera). Control software for the camera head has also been developed.The VP detection and crack detection modules have been evaluated on images captured from library videos of sewer surveys. The results show that the routines successfully locate the VP and can successfully detect coaxial cracks in a predefined region of interest in an image. The AVS as a whole has been tested in a laboratory setting using a short section of concrete pipe and simulated cracks in its wall. The AVS successfully implements a control cycle which determines and fixes the pipe VP, detects coaxial cracks in the pipe wall, orients the camera to attend to those cracks, and then re-fixes the VP.