Automatic extraction of highway light poles and towers from mobile LiDAR data

Mobile LiDAR has been recently demonstrated as a viable technique for pole-like object detection and classification. Despite that a desirable accuracy (around 80%) has been reported in the existing studies, majority of them were presented in the street level with relatively flat ground and very few of them addressed how to extract the entire pole structure from the ground or curb surface. Therefore, this paper attempts to fill the research gap by presenting a workflow for automatic extraction of light poles and towers from mobile LiDAR data point cloud, with a particular focus on municipal highway. The data processing workflow includes (1) an automatic ground filtering mechanism to separate aboveground and ground features, (2) an unsupervised clustering algorithm to cluster the aboveground data point cloud, (3) a set of decision rules to identify and classify potential light poles and towers, and (4) a least-squares circle fitting algorithm to fit the circular pole structure so as to remove the ground points. The workflow was tested with a set of mobile LiDAR data collected for a section of highway 401 located in Toronto, Ontario, Canada. The results showed that the proposed method can achieve an over 91% of detection rate for five types of light poles and towers along the study area.