Indoor Floorplan Estimation from 3D point clouds for Scan-to-BIM

Abstract

Societies depend on the unrestricted availability of their infrastructures. Events such as (natural) disasters, emergencies, or even attacks, could threaten their safety and security. Indoors models provide relevant information that could help in this regard. Their floorplans contain key information such as their location, design, and layout. The architecture, engineering and construction (AEC) community work together to create the respective indoor models within the Building Information Modelling (BIM) framework. BIM modelling has recently gotten the attention in the computer vision domain. The 1st international Scan-to-BIM challenge, organised within the CVPR 2021 conference, helped to establish research interest and common goals between the AEC and computer vision community. In this paper, we introduce a method to estimate floorplans from 3D point cloud data by using the Scan-to-BIM dataset. Our work has been developed by using image processing techniques. It does not aim to replace state-of-the-art approaches, which are more elaborate and robust. Instead, it constitutes a non CPU intensive alternative that fairly estimates floorplans for the Scan-to-BIM dataset.