Bathymetric Anomaly Detection Towards Simultaneous Localization and Mapping on Autonomous Underwater Vehicles

Abstract

Autonomous underwater vehicles (AUVs) are uncrewed vehicles that can dive to deep depths or under ice to map the seafloor in the Arctic. Due to the lack of global navigation satellite systems (GNSS) underwater, AUV's rely on inertial navigation to estimate their position. Inertial navigation suffers from unbounded error drift. Simultaneous localization and mapping (SLAM) can be used to correct the AUV's positional estimate by repeatedly observing landmarks in its surrounding terrain. Bathymetry has been used to define landmarks for underwater navigation using feature extraction techniques designed for optical imagery. This thesis describes the development of a novel anomaly detector, `Bathymetric Anomalies from Anti-Motifs' (BAAM), that is purpose-built to detect unique bathymetric landmarks. BAAM exploits known bathymetric motifs (commonly repeated patterns) to detect bathymetric anomalies which can be used as landmarks for SLAM. Bathymetric motifs were extracted from a region of Delaware Bay bathymetry using a 2-D adapted matrix profile algorithm, geometric transformation- and scale-invariant image matrix profile (GTSI-IMP), that was developed in this thesis. The ability to associate landmarks, of BAAM and existing optical feature extraction algorithms, was evaluated using semi-synthetic sonar images of a separate region of Delaware Bay bathymetry. For the conditions used in this research, the BAAM detector combined with the binary robust invariant scalable keypoints (BRISK) descriptor produced more correct matches than many of the optical feature extraction methods. However, the scale-invariant feature transform (SIFT) detector combined with the BRISK descriptor was found to produce the most correct matches in both the noise-free and noisy semi-synthetic sonar images. Despite SIFT-BRISK's ability to produce more correct matches than BAAM-BRISK on these semi-synthetic sonar images, the landmarks identified in the Delaware Bay bathymetry using BAAM were found to be more unique (anomalous) than those identified using SIFT.