Pole-Shaped Object Extraction from LiDAR Point Clouds of Roads

Abstract

Mobile Laser Scanner (MLS) technology enables the acquisition of high-precision three-dimensional lidar point cloud data of roadside infrastructure elements, including traffic signs and street poles. This research presents an innovative and computationally efficient algorithm for the extraction of pole-shaped objects from MLS point clouds. The methodology addresses the computational challenges associated with large-scale point cloud processing through a trajectory-based segmentation and ground point filtering. The algorithm implements geometric descriptors, specifically linearity and verticality to identify potential pole structures. Subsequently, the statistical outlier removal (SOR) filter is applied to refine the candidate pole detection results. The final classification of true pole objects is achieved through the application of ground clearance criteria. The algorithm's effectiveness was validated through empirical testing across three distinct urban and suburban environments, utilizing data acquired from two different MLS systems. The experimental results demonstrated successful performance metrics, achieving mean precision and recall rates of 97.21.