Um ROV de baixo custo como instrumento para pesquisas de algoritmos e métodos aplicáveis a robótica subaquática

Abstract

In the current market, Remotely Operated Submarine Vehicles (ROVs) with features of transmitting images in real time with low latency have a high financial cost, starting in the range of BRL 4,000 reais, this for the basic models, which are also called template input such as (OpenROV Trident Underwater Drone) or (CHASING Dory Underwater Drone). The high value attributed to ROVs for their construction makes research invol ving computer vision for navigation and data collection in an underwater environment difficult among students, professors and educational institutions, especially schools and public universities. This research demonstrates a low-cost equipment for studies involving computer vision, navigation and data collection in an underwater environment, in addition to having carried out experiments using the ROV involving visual odometry algorithms, namely: LSD-SLAM: Large- Scale Direct Monocular SLAM, ORB Slam, and SVO (Semi-Direct Monocular Visual Odometry), keeping accurate weather information during experiments. Throughout this study, it was possible to conclude that the algorithm that obtained the best result in terms of accuracy and precision in a simulated controlled environment was the SVO (Semi-Direct Monocular Visual Odometry) which managed to achieve the best results in 4 four tests of a total of 6 sixes per session demonstrating its superiority in relation to LSD Slam and ORB Slam, which in turn in the overall average achieved the best result in the average standard deviation of about 0.09, indicating that among the others have less variation between one test and another.