Resumo (PT):
Subsea assets need to be regularly inspected, maintained and repaired. These operations are typically performed using a Remotely Operated Vehicle (ROV) controlled by a pilot that sits in a ship. In order to make operations safer and cheaper, it would be interesting to control the ROVs from land, avoiding the need to hire a ship and crew.
As part of these operations, ROVs need to perform high precision actions such as turning valves, which may be hard to perform in this remote setting due to latency. A semi-autonomous vehicle capable of performing high precision tasks could potentiate the transition to fully remote operations, where people stay on land. In order to develop such a system, we need a robust perception model capable of segmenting the assets of interest.
Additionally, it is important to fuse that information with 3D models of those same assets in order to have a spatial perception of the environment. This fusion may be useful to, in the future, plan the necessary actions to interact with the given asset.
The main goal of this work is to implement a model that: 1) segments different subsea assets of interest, such as valves; and 2) fuse the segmentation information with 3D models of those same assets.
Idioma:
Inglês
Nº de páginas:
70