Naval architecture firm Tymor Marine and Edinburgh University have created the tool with support and funding from CENSIS, Scotland’s innovation centre for sensing, imaging and Internet of Things (IoT) technologies. The machine vision tool, powered by deep learning, will automate and more accurately undertake the reading of draught marks on ships.
The consortium said that draught marks – numbers marked in increments on the side of vessels to indicate how much of the ship is submerged – are currently measured and recorded by eye from the quay or a boat, similar to the way they have been for more than two millennia.
Measurements are often open to interpretation – waves, faded markings, lighting, and marine growth are just some of the factors that can lead to different readings being taken from the same vessel. Mariners also must check the marks on both sides of a ship, which can take hours, requires a boat, and involves health and safety risks.
Accurate draught readings are critical for ensuring a ship’s stability, indicating how much cargo it is carrying and what depths it can safely navigate. The readings are also checked by port authorities to ensure vessels are complying with local limits and regulations.
According to the researchers, their technology uses algorithms applied to video recordings of ships to accurately identify where the water line reaches on a ship’s hull.
Tymor Marine and Edinburgh University will continue to develop the technology, with the aim of creating a smartphone app that allows seafarers to record draught marks and upload them to the cloud for real-time readings.
MORE IN RAIL AND MARINE
Rosie Clegg, naval architect at Tymor Marine, said: “We had been trying to develop this technology for some time, but quickly found there was no off-the-shelf software. Through CENSIS, we found the expertise we needed at the University of Edinburgh to develop our own technology and bring innovation to what is, broadly speaking, a traditional industry.”
Clegg added that over the last 12 weeks the team has proved the concept is feasible and will now focus on its different elements, train it with data being captured with each visit to a vessel, and begin taking it to a commercial level.
“We are also exploring the possibility of applying it to drones, which would make the process even safer,” she said.
Dr Hakan Bilen, reader in the School of Informatics at Edinburgh University, said that early AI researchers thought AI would easily solve visual tasks that we do effortlessly, like recognising digits and estimating waterline, but struggle with more complex situations such as playing a game of chess.
“However, the opposite has turned out to be the case and it is the seemingly simple tasks that we are still finessing,” Bilen said.
“The algorithm we have created for Tymor Marine has been built on the recent advances in deep neural networks. The model takes in a video showing a ship’s hull and identifies where digits on the side of a vessel intersects the water line in a variety of different scenarios.
“We are continuing to build the database by introducing more manual annotations for training and also to improve various components in the method, which should only make it more accurate in the future.”
Poll: Should the UK’s railways be renationalised?
I think that a network inclusive of the vehicles on it would make sense. However it remains to be seen if there is any plan for it to be for the...