Deep learning computer vision-based obstacle detection for autonomous boats
Autonomy is transforming most industries, and maritime robotics is no exception. With 90% of goods moved across the world in vessels, unmanned surface vehicles (USVs) present a considerable market opportunity. Small-sized USVs, in particular, present affordable devices for automated inspection of hazardous areas and periodic surveillance of coastal waters and have a strong potential for a wide-spread use. Apart from efficient control and high-level planning, safe and uninterrupted navigation relies on environment perception, in particular on obstacle detection and timely collision avoidance. The inherent dynamics of aquatic environment combined with a large variety of potential obstacles presents considerable challenges to computer vision systems, leaving robust obstacle detection an open research problem. In this talk an overview of recent work in the ViCoS laboratory (https://www.vicos.si/) on camera-based obstacle detection for USVs and the datasets created to facilitate the research in the USV obstacle detection will be given.