Real-time subsea-SLAM system

Subsea Inspection, Maintenance, and Repair (IMR) operations are frequently carried out in the marine industry. These operations often include mapping the current state of the field or specific assets, by acquiring a 3D representation using sensors such as LiDAR or SONAR. In addition to this, during the IMR operations there can also exist some interaction with the field, such as the measurement of morphological features of the assets, for instance biofouling. To achieve this, Remotely Operated Vehicles (ROVs) need to be equipped with costly sensors or appropriate tools to take measurements and interact with the asset. Although, if a real-time 3D representation was possible by only using a monocular camera, the number of missions to map and access the state of the assets could be reduced and simple observational ROVs could be used, significantly reducing costs and the carbon footprint associated with each mission. To enable 3D live mapping during a subsea operation, we propose a system based on Simultaneous Localization and Mapping (SLAM) and Deep Learning that, upon receiving the live video feed, reconstructs the 3D asset that is being inspected. In this paper we present a prototype system that consists of a 3D GUI where the 3D asset is reconstructed in an incremental online fashion by using deep-learning models fused with SLAM algorithms. The final result is a dense point cloud that can be exported for future use.