Annotation rules and classes for semantic segmentation of point clouds for digitalization of existing bridge structures

Abstract: Germanys large stock of existing bridge infrastructure that has to be digitalized as BIM, politically prescribed. Therefore, point cloud data can be uses. Available open-source datasets and different approaches on semantic segmentation are researched. To fill the gap between theoretical research on point cloud data and manual inspection, a list of object-oriented classes for semantic segmentation is proposed. These classes are put in context with entities of the IFC schema and the German ASB-Ing. An application example with the proposed classes is shown on a manually segmented point cloud and on an approach with the RANSAC Shape Detection algorithm.