Real-time Subsea Communication Cable Detection for AUV-based Inspection

This paper proposes a method for visual inspection of subsea communication cables. As principal carrier for data transmission between continents and islands, such cables are major infrastructure assets that are vulnerable to natural or human causes of damage, but are challenging to survey due to the long distances they span and great water depths they are laid in. AUVs can visually map seafloor cables; however, autonomous cable following along routes with relatively high uncertainties and over varying substrate types is challenging. This research develops a cable detection and following algorithm for real-time on-board processing on camera-equipped AUVs. Simulations on existing cable routes demonstrate effective and repeatable cable following despite introducing random misdetections at the rates expected for the suggested machine learning based image classifier. A study on various levels of incorrect cable detection and different vehicle speeds shows that a low level of false positives in the cable detection is crucial for robust cable following, and lower vehicle speeds lead to a higher visual coverage of cable routes, albeit at an increased survey time.