SLAM using natural features in an underwater environment

This paper presents techniques developed to apply the simultaneous localisation and mapping (SLAM) algorithm to an unmanned underwater vehicle operating in an unstructured, natural environment. It is shown that information from on-board sonar and vision sensors can be fused to select and track regions of the environment that may be used as features to estimate the vehicle's motion. Results including vehicle pose estimates and resulting environment models are shown for data acquired at the Great Barrier Reef in Australia.