Open Access Coding/Images for Automated detection of in-pipe defects in CCTV sewer surveys

This document is a report of Milestone MS 16  of the Co-UDlabs project, funded under the European Union’s Horizon 2020 research and innovation programme and under Grant Agreement No 101008626.

The primary method for defect identification and classification is based around using CCTV to collect images and then these are used for subsequent analysis. In the later analysis images are normally manually inspected and defects are classified according to a standard classification. The defect classification methods used in European countries are different, but have a similar structure, which reflects their historical development and is described in EN13508:Part2. The current national defect classification schemes are generally complex involving a large number of defect codes. Considering the various defect classification codes across Europe the number of unique defect codes is now greater than 300. In contrast in Japan the defect coding system contains just 10 defect types. There are a number of academic studies and more recent companies that are developing data-driven classifiers to link observed defects with a defect classification code. These studies have shown some promise but have resulted in collection of large numbers of images.

This report describe the outcome of work to consider what knowledge could be gained from a more straightforward defect classification approach. This report presents a deep-learning based framework for the automated detection of in-pipe defects in closed-circuit television (CCTV) sewer surveys. The framework utilizes the Ultralytics YOLO v8 model for image processing and defect detection. By eliminating the need for manual feature extraction, this approach simplifies the identification of defects that are challenging to extract features from, such as those found in sewer pipes. The report outlines the methodology, demonstration results, and provides recommendations for further work. All the source code is open access and has been developed to a standard to encourage other, especially non-specialists in small companies and utilities to try and investigate whether a more simplified defect classification scheme can provide the knowledge needed to enhance their management of buried sewer assets. All the code is publically available, with sample images and written software support to allow ease of access. The team at Sheffield will continue to develop this open access approach to software development and encourage those that use the code and uploaded images to report on their findings.