Underground sewer pipe system is one important component in civil infrastructure, which is designed to collect and transport sanitary waste water and storm water to appropriate treatment facilities. Despite the importance of sewer pipe systems, there are various issues facing the construction and management of sewer pipe systems. On the one hand, there are a large number of underground sewer pipes in many countries and most of them have a long history of 30-100 years, making them susceptible to various defects such as cracks, tree root intrusion, and deposits. Such defects can cause the deterioration of sewer pipes and hence leads to severe environmental problems, social issues, as well as the influence and threats to residents. Therefore, it is essential to conduct defect detec...[ Read more ]

Underground sewer pipe system is one important component in civil infrastructure, which is designed to collect and transport sanitary waste water and storm water to appropriate treatment facilities. Despite the importance of sewer pipe systems, there are various issues facing the construction and management of sewer pipe systems. On the one hand, there are a large number of underground sewer pipes in many countries and most of them have a long history of 30-100 years, making them susceptible to various defects such as cracks, tree root intrusion, and deposits. Such defects can cause the deterioration of sewer pipes and hence leads to severe environmental problems, social issues, as well as the influence and threats to residents. Therefore, it is essential to conduct defect detection for sewer pipes to identify and localize existing defects, based on which pipe condition assessment can be performed. In the end, corresponding maintenance and management activities can be decided to prevent potential hazards.

Currently, visual inspection technologies such as closed-circuit television (CCTV) are commonly utilized for sewer pipe inspection. However, inspectors are required to observe the videos through the whole on-site inspection process and manually evaluate the pipe conditions from the captured videos or images, which is time-consuming, labour-intensive and inaccurate. In addition, most utility data are stored and managed in 2D formats, and there is no unified standard for the underground utility data, which is inefficient for sharing information, and localizing utility components. Previous studies attempt to improve the inspection efficiency by automating defect identification and measurement using conventional computer vision techniques, which require prior knowledge of images, handcrafted feature extractor and manual design of classifiers. In recent years, deep learning is widely investigated and applied in various areas. The promising performance of deep learning in image classification and object detection indicates its great potentials for automated sewer pipe inspection. Therefore, this research aims to facilitate efficient defect detection, condition assessment and management of sewer pipe systems using computer vision and deep learning techniques.

For defect detection, an automated approach is developed for detecting, i.e. identifying and localizing defects in CCTV images. Images containing common sewer pipe defects are collected from inspection videos and annotated with ground truth labels. A region-based convolutional neural network (R-CNN) model, Faster R-CNN, is applied and different influential factors, such as the dataset size and network hyper-parameters, on the model performance are studied. In addition, as there are several deep learning models for object detection, the performance of three different state-of-the-art models are evaluated and compared for detecting sewer pipe defects. The architectures of the three models are constructed, after which the models are trained using the same annotated dataset in the same training environment. To demonstrate the viability of real-time automated defect detection, a prototype system is developed for detecting root intrusions and deposits, and is evaluated on inspection videos.

For condition assessment, automated defect segmentation is firstly performed by proposing a unified CNN model integrated with conditional random field (CRF). This work combines the advantages of both deep learning model and CRF for improving the defect segmentation accuracy. An innovated CNN model is proposed as DilaSeg based on dilated convolution and multi-scale techniques to improve the feature extraction process. Meanwhile, the inference algorithm of CRF is converted into RNN layers, which are integrated with DilaSeg to form the unified model named DilaSeg-CRF. Experiments demonstrate that our proposed DilaSeg-CRF can improve the segmentation accuracy significantly. Based on the defect segmentation, the defect area in the image can be obtained, after which pipe condition is evaluated. The condition grading system is first proposed with reference to existing condition assessment standards. Then, the inspection information e.g. manhole ID and defect distance along the pipeline is obtained from the CCTV display through text detection and recognition methods, while the area of the pipe cross section is obtained by image processing techniques. In the end, the defect severity and pipe condition are evaluated based on the grading codes.

For utility management, an integrated management framework is proposed based on building information modelling (BIM) and geographic information system (GIS). A common utility data model is developed, based on which schemas of Industry Foundation Classes (IFC) and City Geography Markup Language (CityGML) are extended for underground utilities. An integrated platform is developed based on the BIM-GIS integration, which supports utility management with different functions. The proposed management framework enables 3D visualization of underground utilities, supports efficient utility data sharing as well as facilitates decision-making process for utility maintenance activities.

This research develops approaches for automated defect detection and condition assessment of sewer pipes, which is expected to reduce the labor cost, inspection time and inaccurate assessment. In addition, this research provides an integrated framework for utility management based on a unified utility data model, which contributes to underground utility digitalization and decision-making for maintenance activities.