No abstract

No abstract

Neste trabalho são desenvolvidos dois algoritmos para a detecção automática de derramamentos de óleo usando informações de sensores locais, como imagens de câmeras visíveis. O primeiro método usa Visão Computacional e Processamento de Imagens, com base em operações de segmentação não supervisionadas e limiares de cores. O segundo usa Redes Neurais Convolucionais, em uma abordagem de aprendizado supervisionado. Simulações numéricas ilustram o desempenho de ambas as metodologias de detecção e é apresentada uma comparação entre eles.

The development of oil spill detection technologies up until now have converged to the broad use of satellite-based SAR (Synthetic Aperture Radar). However, those systems present high amounts of false negatives results (oil spills that are not detected as such) and false positives (any other observation misinterpreted as an oil spill). In order to address this issue, this research aims to produce an in-situ autonomous system which will perform oil spill monitoring more efficiently than the current methods applied by the Oil & Gas industry. Autonomy is key to reduce the unnecessary deployment of recovery crews at elevated financial cost and human safety risks on the case of false positives. Notably, the developed system is expected to drastically reduce the number of false negatives, enabling proper environmental safety within oil E&P infrastructure. The proposed autonomous system (called ARIEL) is composed of an Unmanned Surface Vehicle (USV, or autonomous boat) and a Drone (specifically, a multirotor vertical take-off and landing drone). Those subsystems work together to perform a pre-programmed mission given by a human. Both are capable of navigating through a path made of waypoints, limiting a region of interest where oil spill is to be monitored. The main benefits from the configuration of the system being developed is that it has much lower initial and maintenance costs when compared to current solution. Also, it can be easily deployed as a fleet, scaling its coverage and sweep rate. Oil detection sensors are to be installed on both the autonomous vessel and drone, providing a double layer solution for eliminating errors. The first oil spill detection layer runs inside the drone, aiming at minimizing the probability of false negatives. This will be done via a visible and thermal camera package. The second layer runs in the USV via a fluorsensor. Firstly, it validates all possible incidents reported by the drone. Secondly, it double-checks all information gathered by the drone to check any oil spill incident that went unnoticed. Furthermore, the USV is to be equipped with metocean sensors for adding the capability of tracking the oil slick movement and predicting its position in time. Measurements provided by wind and current sensors will be used to guide monitoring missions and can potentially be used to plan recovery operations. Experimental analysis on both controlled environments (laboratorial and sheltered waters) and operational environment (open waters) are under execution. Not only results on sensory data processing but also the challenges overcome on the joint task of landing a drone on a moving vessel are presented is this work.

No abstract