Extraction of uban buildings from LiDAR point clouds using DBSCAN clustering

Abstract

Airborne Light Detection and Ranging (LiDAR) datasets are widely used for urban analysis and three-dimensional modelling due to their high spatial accuracy and ability to represent complex forms. In many LiDAR datasets, points are already classified into thematic classes such as ground, vegetation, and buildings. However, even when building points are provided as a separate class, individual buildings remain merged within a single point cloud and require additional processing to be separated into distinct objects. This study presents a simple method for separating individual buildings from a building-class LiDAR point cloud. It applies the density-based clustering algorithm DBSCAN (Density-Based Spatial Clustering of Applications with Noise) to automatically group spatially connected points into clusters representing individual buildings. The method does not require the predefined number of buildings and is robust to noise and variations in building geometry. The resulting clusters represent separate building structures that can be exported as independent datasets for further processing. Such segmented building point clouds can serve as an initial step for applications including three-dimensional city model generation, urban analysis, and integration with geographic information systems.