Enhancing maritime safety: developing an accessible real-time semantic wave imaging analyser for seakeeping

Our project aims to predict wave-vessel-payload interactions by extracting semantic wave feature in real time. This requires accurate 3D wave geometry information over time – a dynamic 4D scene. Our methodology takes advantage of computer vision, deep learning, computational mechanics and human biomechanics approaches to correlate visual wave features to hydrodynamic and biomechanical loads acting on the vessel, payload and crew.

We propose the following specific objectives.

1. Acquire and create accurate dense stereo vision based 4D surface wave
reconstruction onboard vessels as an essential component to evaluate the accuracy of a new sparse wave model.

2. Develop real-time visual hydrodynamic wave feature extraction and interpretation algorithms by comparing different computer vision and deep learning approaches.

3. Correlate hydrodynamic loads of new real-time wave models with vessel dynamics, and predict dynamic loads experienced by the payload and crew.

We intend to make our research outcome accessible to academic and industrial research groups by creating a comprehensive tutorial website of the open-source wave analyser, stereo database and the low-cost imaging hardware.

Our proposed research aims to develop an ocean wave imaging analyser to predict wave-vessel-payload-crew interaction. This is a currently missing prerequisite for optimal seakeeping of fast vessels. Seakeeping, concerning the control of vessel motion when subjected to waves and the resulting effects on humans, systems, and mission capacity, remains one of the biggest challenges in maritime safety. Vessel operational practices (48%) and human factors (17%), both key to seakeeping, have been the main safety recommendations amongst 1212 investigations, conducted by the European Maritime Safety Agency in the past decade.

Before making any control decisions, mitigating detrimental effects on seakeeping requires accurate and real-time modelling of the approaching waves in the perimeter of the vessel. The predicted wave loading is essential for any precise estimation of vessel motion, but it is absent. To derive such a model, 3D wave geometry evolving in real time – a dynamic 4D scene – is required. However, the computational time required for existing sensing and modelling approaches are too long for the decision windows of any vessel operations. This process presently takes more than tens of seconds in order to anticipate and react at close proximity. This leads to the three specific challenges we propose to tackle.