Monocular ORB-SLAM Application in Underwater Scenarios

Abstract

This paper presents an experimental evaluation of monocular ORB-SLAM applied to underwater scenarios. It is investigated as an alternative SLAM method with minimal instumentation compared to other approaches that integrate different sensors such as inertial and acoustic sensors. ORB-SLAM creates a 3D map based on image frames and estimates the position of the robot by using a feature-based front-end and a graph-based back-end. The performance of ORB-SLAM is evaluated through experiments in different settings with varying lighting, visibility and water dynamics. Results show good performance given the right conditions and demonstrate that ORB-SLAM can work well in the underwater environment. Based on our findings the paper outlines possible enhancements which should further improve on the algorithms performance.