Efficient visual SLAM using selective image registration for autonomous inspection of underwater structures

Visual inspection of underwater structures including ship-hull inspection has been performed by human divers. It is a highly dangerous task and thus can be a potential application for unmanned underwater vehicles. This paper introduces an efficient visual simultaneous localization and mapping (SLAM) algorithm that can be applied to the autonomous inspection of underwater structures. Considering that visual features are sparsely located on the surface of typical underwater structures, the proposed visual SLAM algorithm employs a selective image registration scheme consisting of key-frame selection and key-pair selection. By using only potentially effective images and image pairs for feature-based image registration, the computational burden of the visual SLAM can be substantially reduced, compared with the conventional method. Experimental results using a hover-capable unmanned underwater vehicle verify the practical feasibility and performance of the proposed methodology.