Authors: Islam, Md Jahidul, {Quattrini Li}, Alberto , Girdhar, Yogesh A, Rekleitis, Ioannis
Abstract: Computer Vision has surpassed many impressive milestones in recent years in the field of robotics and Artificial Intelligence (AI). Cameras are passive, inexpensive sensors that produce big volumes of data; as visual sensing is one of the primary perception modalities for humans, it is also human interpretable. However, the underwater realm is a challenging domain with severe light attenuation where visual perception techniques that work well above water often fail. Nevertheless, several application areas are looking into vision to obtain semantic as well as metric information. Subsea exploration, envi-ronmental monitoring, oceanography, aquaculture, marine archaeology, search and rescue, submarine structure inspection, and resource utilization are some of the major areas where computer vision has been applied with great success. Autonomous Underwater Vehicles (AUVs) and Remotely Operated Vehicles (ROVs) are generally deployed in these applications for autonomous or semiautonomous missions, often alongside human divers. Unlike terrestrial robots, ensuring robust visual guidance for underwater robots faces unique challenges due to the domain-specific operational complexities associated with underwater sensing and estimation. In this chapter, we present the vibrant scientific literature that attempts to address these challenges for robust underwater visual perception, and then elaborately discuss the emerging vision technologies of subsea robotics applications.