Vision-Based Informative Path Planning for Estimating Oil Spill Area Using Multiple Drones

Oceanic oil spills have a devastating impact on marine life. It is important to implement effective and accurate monitoring solutions to minimize environmental damage. This paper introduces a vision-based informative path planning strategy that employs multiple drones to autonomously estimate the oil spill area. A Bayesian Optimization (BO) framework is utilized to guide the movement of drones, in which a Gaussian Process is used to predict a model of the oil spill based on the collected images in real time. In addition, an acquisition function is developed to guide UAVs to the most informative regions. The paper compared various kernel functions and acquisition functions, selecting the most efficient combination for oil spill monitoring. The proposed algorithm was evaluated in three different image-based test fields. The results demonstrate the drones focus on areas with more oil spill and at the same time achieve small error oil spill area prediction rates. Future work aims to extend this framework to high-fidelity simulation and experimental environments to test algorithm operation under more realistic and dynamic conditions.