3D building internal structural component segmentation from point cloud data using DBscan and modified ransac with normal deviation conditions

Abstract

Nowadays, the laser scanner plays an important role as a tool to capture building structure in a form of point cloud which can be used to draw a blueprint or a floor plan for reconstruction or renovation of an existing building. The point cloud data represent a shape or an object in three dimensions. This point cloud together with the utilization of Building Information Modeling (BIM) which is a workflow that provides information about the foundation and structure measurement of a building, make a planning and design to construction more convenient and efficient in several aspects than the traditional way. However, using BIM to design or construct the internal structure of an existing building from point cloud manually can be very time consuming and requires a highly skilled engineer. The development of a method that can automatically construct the internal structure of an existing building from point cloud can facilitate the process of model creation. Among the whole process of construction, segmentation is an important process that identifies the main components of a building, which is still a challenging problem for researchers to develop an algorithm that can automatically segment the internal structural components of the building. The objective of this research is to develop a segmentation method that can extract the planar structures of a building from point cloud using the modified Random Sample Consensus (RANSAC) algorithm. The original RANSAC is modified by reducing the computational complexity using localized sampling, and the quality of segmentation is improved by adding normal deviation with connectivity constraints to RANSAC’s score calculation. The original RANSAC and the proposed method were evaluated on the International Society for Photogrammetry and Remote Sensing (ISPRS) benchmark dataset. The segmentation results of the proposed method were visually compared to the results of the original RANSAC and it can be seen that the extracted components from the proposed method are closer to the actual structure and can also preserve the overall characteristics of the buildings.