Automated Defect-Detection System for Water Pipelines Based on CCTV Inspection Videos of Autonomous Robotic Platforms

Abstract

Water is an essential element for the survival of human beings and a nation. Nowadays, water utilities perform regular inspections of the internal conditions of the pipelines via autonomous robotic platforms. The human operator then analyzes the recording of the platforms to identify the defects inside the water pipelines. This manual assessment process is often time-consuming, exhaustive, and error-prone. Hence, this article proposes an automated defect-detection framework channel-spatial attention Mask-Canny-regional convolutional neural network (CSA-MaskC-RCNN) that automatically detects and classifies defects. An adaptive CSA mechanism algorithm is proposed to extract the detected features more efficiently. The features are then passed through a modified defect detector, MaskC-RCNN, to classify the defects on the videos obtained through the autonomous robotic platforms' closed-circuit television cameras. Then, the trained model is employed to develop a defect-detection unit interface tool to perform the defect assessment. The results from our study show that our model can outperform the state-of-the-art model with a mean average precision of 86.89%. Hence, integrating this automatic defect-detection system can save time and cost for the human operator and aid them in making timely decisions for pipe repair/rehabilitation.