Floor plan reconstruction from indoor 3D point clouds using iterative RANSAC line segmentation

Abstract

Indoor Floor Plans (IFPs) are crucial in many fields, such as architectural design, BIM generation, indoor navigation, as well as reliability analysis and safety assessment in civil engineering. The traditional manual creation of IFPs is labor-intensive and time-consuming, hence the growing interest in their automatic generation. To automate IFP creation and ensure their consistency with original indoor layouts, a method for generating IFPs from indoor 3D point clouds is proposed. This method uses a complete point cloud from an indoor scene as the sole input and produces a corresponding vectorized IFP. It first identifies wall points from the 3D point cloud, and projects them onto a 2D plane. An iterative RANSAC-based clustering algorithm is then used to detect and segment line segments from the 2D point cloud. Finally, we restore the topological structure of the discrete line segment set and outputs a vectorized IFP composed of continuous line segments. When tested on two different scale indoor point cloud datasets, this new method achieved over 90 % of accuracy in line segment reconstruction and 97 % of IoU in IFPs reconstruction, proving its effectiveness in non-Manhattan indoor scenes. Compared with other representative methods, Floor-sp and ASIP, this new method offers more detail and higher accuracy.