Semantic Segmentation of Building Point Clouds Using Deep Learning: A Method for Creating Training Data Using BIM to Point Cloud Label Transfer

Creating deep learning classifiers requires large labeled datasets; and creating large labeled datasets requires elaborate crowdsourcing systems and many hours of manual human effort applied to classification and data entry. Fortunately, much of this effort can be bypassed in the building industry because of as-built building information models ( BIMs), a semantically rich form of facility information. From these BIMs, semantics can be transferred to point clouds. This paper presents a method for creating large labeled datasets for training deep neural networks to semantically segment point clouds of buildings. Geometry and attached semantics are extracted from a BIM. The geometry is registered with the point cloud and the BIM semantics are copied to the points in the point cloud. The presented method enables organizations with access to as-built BIMs to forgo the effort of creating large labeled datasets and instead use the embodied effort in their pre-existing BIMs.