Sistema computacional integrado a modelos digitais para reconstrução de estruturas subaquáticas com alta definição

Abstract

The development of research focused on the inspection and scanning of subsea equipment is becoming increasingly important as oil and gas exploration evolves in deeper waters, in addition, there are several other demands for the use of robots and autonomous sys tems, such as underwater archaeological exploration, ocean monitoring, marine biology and geological studies. Frequently, the lack of accurate geometric information on the subsea equipment imposes difficulties in performing operations related to inspection and maintenance. A challenge presented is the geometrically accurate reconstitution of the environment where these subsea equipment are located. From studies on underwater 3D scanning, an evolution in methods is observed with the aim of improving the efficiency of this type of system, but new research efforts are needed, since these optical systems are complex. Point clouds formed from optical scanners contain noise and outliers, which hinder the scene reconstruction process. This justifies the interest in developing new te chnological solutions to overcome the difficulties in obtaining information through the use of optical acquisition methods. This thesis seeks to consolidate point clouds cluttered and provide an accurate three-dimensional representation of underwater equipment structu res. For this purpose, a new data fusion method of merging software based on Structure from Motion - SfM and Multiview Stereo is developed with the goal of minimizing errors introduced into the point clouds as a result of motion disturbances in the operating en vironment. The results achieved during the deep-water tests highlight that the system presents good results, with measurement errors between 0.47 and 4.12 millimeters for me asurement distances between 5 to 6 meters, and an uncertainty between +/- 0.81 and +/- 6.50 milimeters. The feasibility of implementing this system is confirmed by the digitally reconstructed equipment maps obtained during the testing phase. Therefore, this thesis brings a new and relevant contribution to the 3D reconstruction of these structures with the dimensional precision necessary for robotic operations and considering the inherent issues imposed by this environment.