Development, Evaluation and Validation of a Stereo Camera Underwater SLAM Algorithm

Abstract

In this work the development of an algorithm for visual underwater localization is described. It spans the complete process from the initial idea, the development of a suitable underwater vehicle for testing to the algorithm's experimental validation in real underwater environments. Besides the development and validation of the visual SLAM algorithm, the methodology for its evaluation is a key aspect of this work. The resulting SURE-SLAM algorithm uses a stereo camera system and basic vehicle sensors (AHRS, DPS) to compute a complete, error-bounded localization solution for underwater vehicles in real-time with similar quality as state-of-the-art acoustically stabilized dead-reckoning approaches. The robustness of the algorithm as well as its limitations and failure-cases are established by extensive field testing with the AUV Dagon, which was developed during this thesis as test and evaluation vehicle.