Authors: Maarten Bassier, Maarten Vergauwen, Bjorn Van Genechten
With the increasing popularity of Building Information modelling (BIM), the demand for accurate as-built models of existing buildings is rising. However, the manual creation of these models is labour intensive and error prone. Therefore, automation of the process is a must. One of the key factors in the automated Scan-to-BIM process is the labelling of the data for further reconstruction. Currently, semantic labelling is still ongoing research. This paper presents a flexible method to automatically label highly cluttered vector models of existing buildings. In our proposed method, a reasoning framework is used that exploits geometric and contextual information. A major advantage to our approach is that our algorithm can label both cluttered environments and large data sets very efficiently. Unlike other solutions, this allows us to label entire buildings at once. In addition, the implementation of our algorithm and the platform we use allows for flexible data processing, visualisation of the results and improvement of the labelling process. Our work covers the entire labelling phase and allows the user to label data sets with a minimal amount of effort.
Keywords: Semantic labelling; Scan-to-BIM; Vector model; Building modelling