Point Cloud Data Extraction Combining Surface Variation and Ransac for Digital Archiving of Shell Structures

Abstract

The importance of digitally archiving existing architectures for cultural and research purposes is continuously increasing. Therefore, there are growing needs to extract significant data and reduce the amount of data in 3D measurements. Point clouds obtained from 3D surveys of entire buildings are expected to be used as a digital archive for various applications, such as structural analysis and renovation planning. However, effectively managing the vast amount of data from these surveys remains a challenge. Building point-clouds are typically massive and often contain unnecessary points and missing parts. This paper presents a novel method that combines surface variation and Random Sample Consensus (RANSAC) for determining the shape parameters of point clouds with curved surfaces. Our approach is efficient, not only in reconstructing surfaces of individual shells but also in handling combined and interpenetrated ones. The approach is also effective in processing the point cloud from actual measurement, demonstrating resilience to outliers and missing parts. Additionally, when applied to the measured point cloud for an example structure composed of seven intersecting domes, our approach accurately estimates the surface with an error margin of 0.06%–0.24%. Significant reduction in data volume, from 6.69 GB to 33 KB, is also achieved.