Water Journal : Water Journal May 2012
pipeline cleaning & maintenance refereed paper technical features 58 MAY 2012 water Abstract A considerable amount of research into automated or semi-automated interpretation of images of sewer or other pipes has been carried out over the past two decades. The Commonwealth Scientific and Industrial Research Organisation (CSIRO) has had significant involvement in this research almost from its outset. A major early project in this area was the PIRAT project, which was a collaborative effort involving two divisions of CSIRO and Melbourne Water. In this project an image acquisition robot was built, and software for the automatic interpretation of pipe images using computer vision and other artificial intelligence techniques was developed. Recent CSIRO work in the area has been focused on image interpretation, where reliance is made on commercial image acquisition systems. Techniques for pipe defect and pipe feature detection have been developed and, through university partners, techniques for 3D pipe reconstruction and large scale processing of pipe image collections by neural self- organising maps have been developed. Introduction The process of inspection of the interior of buried pipes is made up of two parts. These are the image acquisition component and the image interpretation component. The image acquisition system is usually made up of a robotic device that moves through the pipe obtaining images. The images may be greyscale, colour, range or other sensor modality. In current pipe inspection practice the interpretation component is manual. This manual image interpretation is a time-consuming, labour- intensive and error-prone operation. Automation has the potential to increase its objectivity and efficiency. Such an automated interpretation system processes images obtained by the image acquisition device with the aim of identifying defects and features and, possibly, providing input into a system that produces a report about the state of the pipe. It may be fully automated or semi- automated, in which case it operates in conjunction with a human operator. One possible function of a semi-automated system is to provide a screening system to filter out images of pipe sections that are definitely non- defective, leaving potentially defective areas to be assessed by the human operator. Such a screening operation has the potential to increase the efficiency of the image interpretation process. An early semi-automatic inspection system was the German KARO system (Kuntze et al., 1994), which consisted of a multi-sensor inspection device and a two- pass interpretation system. During the largely automatic first pass, a hierarchical fuzzy logic sensor fusion algorithm was used to identify candidate defects that were then investigated in detail by the operator during a second pass. The PIRAT system (Kirkham et al., 2000) was developed shortly after the KARO system. It was made up of an inspection device providing range images (obtained by laser striping) together with an interpretation system using neural networks and other AI techniques (Mashford, 1995). The image acquisition device for the PIRAT system is shown in Figure 1. Our recent work has focused on the development of an interpretation system rather than on the building of an inspection device. Data for such an interpretation system could be provided by a commercial inspection system such as that shown in Figure 2. Such an inspection system will move through a pipe and obtain a number of colour images that may then be combined to form an unwrapped pipe image. The images obtained by the inspection system, or the unwrapped pipe image generated from these images, are processed by the image interpretation system. The accuracy of the image interpretation process would depend on the quality of the images obtained by the inspection device. It may be difficult to obtain satisfactory accuracy in the case of sewer pipes that have not been pre-cleaned. J Mashford, D Marney, S Burn A review of current status for automatic recognition of pipe defects from a robot CSIRO RESEARCH INTO COMPUTER AIDED IMAGE INTERPRETATION FOR AUTOMATIC PIPE INSPECTION Figure 1. The PIRAT image acquisition device. Figure 2. A commercial image acquisition device.
Water Journal July 2012
Water Journal April 2012