Automatic point cloud segmentation using RANSAC and DBSCAN algorithm for indoor model

Abstract

Indoor modeling is a crucial aspect of architecture, engineering, and construction (AEC) projects. While terrestrial laser scanners (TLS) are the most popular tool for acquiring indoor geometry, processing point clouds from TLS scans with manual methods can be inefficient and error-prone. This study proposes a machine learning algorithm to automatically segment point clouds acquired by low-cost TLS. Random sample consensus (RANSAC), a simple yet effective algorithm for segmenting planar surfaces such as walls, ceilings, and floors, is used in the segmentation process. The resulting segmentation is then refined using density-based spatial clustering of application with noise (DBSCAN) to group nearby points that were not segmented correctly by RANSAC into the appropriate segment. The result is a segmented point cloud consisting of five indoor elements: wall, ceiling, floor, column, and interior. The algorithms were found to be effective for segmenting small and simple rooms. For larger or more complex rooms, segmentation can be performed by dividing the room into several parts and applying the algorithms to each partition. Overall, the study demonstrates the potential of machine learning algorithms for automating point cloud segmentation tasks in indoor modeling, especially for low-cost TLS scans.