Dynamic Infrastructure and University of Bath Test AI for Drainage Inspections

Dynamic Infrastructure and the University of Bath are testing how AI methods detect blockages and obstructions in drainage inspection images. The organizations said in a press release that the research also measures how models perform on sites not included in their training data.

The study compares vision language models, vision foundation models, supervised deep learning and classical machine learning methods on the same task. It assesses accuracy, robustness, labeled data requirements and the ability to transfer between sites with different construction, camera equipment and inspection practices.

Early results show performance varies across both methods and sites. Current vision language models did not provide the robustness required for reliable use on this specialized task, while more targeted methods performed better. The partners plan to continue the research and develop it into a peer reviewed publication.