Automatic Detection of 3D Steel Structures in As-built Point Clouds

Rogério Pinheiro de Souza,Paulo Ivson Netto Santos,Cesar A. Sierra Franco,Alberto Barbosa Raposo

DOI: https://doi.org/10.1145/3373509.3373565

2019-10-23

Abstract:The use of Building Information Modeling (BIM) is a growing reality in the civil industry. Its key information is the 3D geometric model of the building. With the need to apply such methodology to existing buildings, researchers have focused on how to automatically generate 3D geometric information from point clouds provided by laser scanning sensors. Most of this research has focused on the recognition of either planar (e.g., walls) or cylindrical (e.g., piping) structures. Few works have dealt with the detection of structural steel elements, due to its particular shape, and only recently focused on detecting their dimensions. However, in these approaches the point cloud of each structural element was manually separated from the point cloud of the entire building. This creates a challenge since the manual segmentation of a point cloud is a long and subjective process. In this paper, we propose a novel method that automatically detects structural steel elements in a 3D point cloud of existing buildings, without the need for previous segmentation. The proposed technique combines a slice approach with image processing and a machine learning strategy. Performance results show the effectiveness of our approach to detect the desired elements with more than 94% of precision. These results open new possibilities to create automated pipelines to generate accurate 3D BIM models.