Underwater visual acoustic SLAM with sensor calibration

Abstract

Underwater environments present significant challenges for visual Simultaneous Localization and Mapping (SLAM) systems due to limited visibility, poor illumination, and the sporadic loss of structural features in images. To address these issues, this thesis focuses on underwater SLAM by fusing data from a Doppler Velocity Log (DVL), stereo cameras, and Inertial Measurement Unit (IMU) within a graph optimization framework. Additionally, we propose an extrinsic and DVL transducer calibration algorithm that estimates the extrinsic parameters and DVL transducer orientation without requiring additional setups. Furthermore, a novel observability-aware, entropy-based active calibration method is introduced to actively search for the next optimal poses for reliable calibration. Extensive experimental results demonstrate that our proposed SLAM system surpasses state-of-the-art underwater and visual-inertial SLAM systems in terms of localization accuracy and robustness, and highlight the precision of our calibration algorithm.