Towards UAV-Based Absolute Hierarchical Localization in Confined Spaces

Abstract

To perform safe, autonomous, visual inspections of maritime vessels, accurate and lightweight localization of UAVs in dark and confined areas is required. Such visual inspections are performed by expert surveyors, periodically, and aim at the classification of the vessel by identifying defect locations within the frames of the ship. The paper at hand presents our initial efforts in absolute localization as a solution to the aforementioned problems using a UAV. Employing a combination of Visual Inertial Odmetry and detection of existing known structural 3D landmarks, an absolute and hierarchical localization system is created. With this approach there is no need for detailed maps of the environment, but only high level information -i.e. few landmark points- from the structural drawings of the ship. The system is tested within a mock-up model of a topside water ballast tank and its localization is compared to a groundtruth, which is estimated based on ArUco markers mounted on the walls inside the model. Our initial results indicate that using geometric landmarks, extracted from vessel designs, greatly benefit the UAV localization accuracy as well as the localization of defects with regards to the vessel body frame.