Object recognition by machine vision to enhance scene interpretation in an underwater application

Collaborative work is going on between University College London and Sira Ltd. to introduce artificial intelligence in the vision system for an underwater remotely operated vehicle (ROV). Onboard vision systems act as a close-range navigating tool for docking the vehicle onto a remote structure, as well as allowing close-range visual inspection. As the underwater environment produces poor quality pictures, the real-world aim is to utilise object recognition by the machine vision system to aid scene interpretation for the operator on the surface. Multiple independent camera views are analysed for object recognition. The extracted features are matched to a computer 3D world model, incorporating a-priori knowledge of the surroundings. Intelligence is required to filter relevant data and upgrade the computer model as necessary. The complex process of data fusion from multiple sensors is addressed to improve object recognition and correlation between the real world and the computer model. The visual interface to the operator is to be enhanced using the clearer computer model, enabling a more precise position of the ROV to be determined.