Multisensor fusion for marine infrastructures’ inspection and safety

Abstract

Periodic inspection and maintenance of marine infrastructures are essential to ensure the safety and uninterrupted operation. Accurate localization and pose estimation of the remotely operating underwater vehicle is of prime importance to conduct any robotic marine intervention. A generic visual-inertial registration framework is proposed here to estimate the drift-free rotational and translational pose of the underwater vehicle that can be directly fed to the high level vehicle controller. Underwater image formation model based image enhancement technique along with the pose estimation module is the key of the proposed G-VINS algorithm. The theoretical formulation along with the experimental results shows the robustness and the applicability of the proposed method in marine environment characterized by different level of turbidity.