A team from MTRU are working with Trinity House to investigate aids to marine navigation (buoys to me and you), specifically how easy they are to detect and to determine what class of buoy it is. For the land lubbers amongst you, navigation buoys have different colours and shapes at the top that indicate something about the waterway that should be observed as you sail or motor around. For example, red and green markers indicate the approach to harbours and yellow and black markers indicate underwater obstructions and which side you should safely pass.
Understanding these meanings is standard learning for sailors and efforts have been made to see how good computers are at this task. ShorelineNet is a project that investigates deep learning models to detect obstacles you might encounter on the water, including buoys. Our investigation focuses on identifying navigation buoys and determining their class.
To train a machine learning model you need lots of data and although there are lots of buoys on waterways around the UK, there are not many good pictures of them that we can use. So we’re using synthetic data to boost training of the models and to explore features of the buoys to inform future designs of navigational aids.
It’s very exciting to be part of this project with Trinity House that could have a significant impact on improving safety at sea.