Have you ever wondered how many open-source perception datasets are available for autonomous vessels—and how this number compares to the booming field of autonomous driving?
A new study by the University of Antwerp, a partner in the Inno2mare project, published in Ocean Engineering, explores exactly that.
In this comprehensive work, the researchers present a systematic survey of open-source datasets and computer vision techniques used in autonomous vessel navigation. The paper provides an in-depth overview of research developments from the past decade, serving as a valuable reference for both newcomers and experienced researchers in the field of maritime autonomy.
The Rise of Unmanned Surface Vehicles (USVs)
Unmanned Surface Vehicles (USVs) have emerged as key players in modern maritime operations, enabling a wide array of tasks in challenging environments—without the need for onboard crew. With their increasing adoption, the importance of vision-based capabilities, such as object detection and scene segmentation, has grown significantly.
To build reliable vision algorithms, datasets are essential. They enable model training, testing, and validation under realistic maritime conditions. As a result, numerous recent studies have focused on developing and releasing vision datasets tailored specifically for USVs.
At the same time, various deep learning methods, particularly in computer vision are, being applied to enhance USV navigation in general and particularly in our pilot project 3. However, until now, there has been no unified review that combines insights from both datasets and the AI techniques applied to them.
This Study Delivers:
- A comprehensive review of publicly available USV vision datasets
- A survey of deep learning techniques applied to USV perception
- A detailed analysis of current limitations, challenges, and trends
- Insights into opportunities for future research and development
As the field of autonomous maritime systems evolves, this work provides a much-needed foundation to guide future innovations in USV vision and navigation. It underscores the importance of open-source datasets and highlights areas where the research community can collaborate to push the boundaries of maritime autonomy.
Read the full study in Ocean Engineering and stay tuned for more updates from the Inno2mare project:https://doi.org/10.1016/j.oceaneng.2025.121501
