PinSout: Automatic 3D Indoor Space Construction from Point Clouds with Deep Learning

Abstract

With the development of Light Detection and Ranging (LiDAR) technology, point cloud data is a valuable resource to build three-dimensional (3D) models of digital twins. The geospatial 3D model is the principal element to abstract a geographic feature with geometric and semantic properties. The 3D model data provides more efficiency to handle, retrieve, exchange, and visualize geographic features compared to point clouds. However, the construction of 3D models, especially indoor space where various objects exist, usually necessitates expensive time and manual labor resources to organize and extract the geometry information by authoring tools. This demonstration introduces Point-in Space-out (PinSout), a new framework to automatically generate 3D space models from raw 3D point cloud data by leveraging three open-source software: PointNet, Point Cloud Library (PCL), and 3D City Database (3DCityDB). The framework performs the semantic segmentation by PointNet, a deep learning algorithm for the point cloud, to assign a target label to each point from a point cloud, such as walls, floors, and ceilings. It then divides the point cloud into each label cluster and computes surface elements by PCL. Each surface is stored into a 3DCityDB database to export an OGC CityGML data. Finally, we evaluate the accuracy with two datasets: a synthetic point-cloud set of a 3D model and a real dataset taken from the exhibition halls.