Vision Based Automated Pipeline Inspection

Most nations today have a highly complex and old pipeline network, used for transporting gas, water, telecommunication cables, sewage, etc., that demands regular inspection and repairs. Majority of these pipelines are buried underground for aesthetic purpose, that makes regular maintenance work more costly and difficult. Manual CCTV inspection is a widely adopted inspection method that relies on an operator controlling the forward-looking, CCTV mounted mobile robot and reviewing lengthy CCTV footage. The automation of such a subjective and tiring process is crucial. With recent advances in the area of the neural networks, it is possible to develop automated systems capable of inspection using the CCTV images. The Sewer-ML data set, containing 1.3 million real images real pipeline inspection images, was selected for this study. A smaller subset of the complete data set was used to identify the best-performing deep learning model. Two different approaches, i.e., Hierarchical and Unified Defect Classification approaches were systematically evaluated. Our method using hierarchical defect classification with Mobilenet and Densenet-169 resulted in overall <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$F2_{CIW}$</tex> score of 63.22% and 57.68% and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$F1_{Normal}$</tex> score of 90.2% and 89.4% on validation and test split of the pruned data set.