Dense, Sonar-Based Reconstruction of Underwater Scenes
Pedro Vaz Teixeira, Ph.D., 2019
John Leonard, Advisor
The accuracy of a mapping platform is determined by the accuracy of the mapping sensor measurements and of the pose and sensor offset estimates. Surface-based surveying platforms combine highly accurate global navigation satellite system (GNSS) receivers with attitude and heading reference systems to instrument mapping sensor pose. Rapid attenuation of electromagnetic signals precludes the use of GNSS underwater. The lower accuracy and operational limitations of acoustic positioning systems shifts the accuracy burden to aided inertial navigation systems (INS). As uncertainty in the pose estimates of an aided INS grows unbounded over time, use of simultaneous localization and mapping (SLAM) techniques becomes necessary.
This dissertation presents techniques aimed at improving the accuracy of maps produced using multibeam sonar. First, we propose robust methods to obtain accurate range measurements from sonar data in the presence of noise, sensor artifacts, and outliers. Second, we propose a volumetric, submap-based SLAM technique that leverages map information to correct for drift in the mapping platform’s pose estimate. Third, we propose a dense approach to the sonar-based reconstruction problem, where the pose estimation, sonar segmentation and model optimization problems are tackled simultaneously under the factor graphs framework. Finally, we provide experimental results that validate the proposed techniques.