3D anomaly inspection system for sewer pipes using stereo vision and novel image processing

This paper describes and evaluates a novel 3D inspection system to detect anomalies in sewer pipes using stereo vision coupled with novel image processing algorithms. Currently, most commercial pipe inspection systems are designed with one or more Closed-circuit Television (CCTV) cameras. These systems are slow, costly and have limited accuracy (caused by human and environmental factors). More sophisticated systems (Laser-based, Infrared Thermography, Ultrasonic-based and Ground Penetrating Radar) suffer from: low resolution, high noise, high operational costs and an inability to detect water infiltration. The main objective of this research is to apply stereo vision and robust image processing to generate 3D images of anomalies in sewer pipes in order to achieve high efficiency and accuracy for pipe inspection. The results show that various types of defects are successfully detectable. In addition, the correspondence time can be reduced by up to 45% and the accuracy of disparity maps is maintained compared to traditional local correspondence algorithms. Each component of the proposed system was tested individually with real and simulated data sets.