Autonomous Surface Vessel Navigation using ORB-SLAM 3 with Shoreline Assistance

Full autonomous surface vessels (ASVs) require autonomous navigation algorithms. Global Navigation Satellite Systems (GNSS) are a key component of navigation algorithms, providing reliable positioning information for the vessel. However, the reliability of GNSS depends on sky-visibility, which is not always reliable in near-shore regions, especially in riverine environments. Such regions also present dangers to ASVs due to the proximity of non-aquatic obstacles. Simultaneous localization and mapping (SLAM) algorithms provide a solution in riverine and near-shore regions, taking advantage of the abundance of visual data as well as inertial measurements. Algorithms like ORB-SLAM 3 can function well in such situations, but are adversely affected by unwanted features reflected by the moving water surface. This work introduces a method of improving ORB-SLAM 3 and other feature-based SLAM algorithms by water surface segmentation in order to achieve reliable ASV navigation in riverine conditions where GNSS is unreliable. This produces a 10-20% improvement in like-for-like performance measurement based on generated trajectory information.