Abstract
The number of light poles and their position (in terms of density and offset off the roadside) have significant impacts on the safe operation of highways. In current practice, inventory of such information is performed in periodic site visits, which are tedious and time consuming. This makes inventory and health monitoring of poles at a network level extremely challenging. To relieve the burden associated with manual inventory of poles, this paper proposes a novel algorithm which can automatically obtain such information from remotely sensing data. The proposed algorithm works by first tiling point cloud data collected using light detection and ranging (LiDAR) technology into manageable data tiles of fixed dimensions. The data are voxelized and attributes for each data voxel are calculated to classify them into ground and nonground points. Connected components labeling is then used to perform 3D clustering of the data voxels. Further clustering is performed using a density-based clustering to combine connected components of the same object. The final step involves classifying different objects into poles and non-poles based on a set of decision rules related to the geometric properties of the clusters. The proposed algorithm was tested on a 4 km rural highway segment in Alberta, Canada, which had substantial variation in its vertical alignment. The algorithm was accurate in detecting nonground objects, including poles. Moreover, the results also highlight the importance of considering the length of the highway and its terrain when detecting nonground objects from LiDAR.
Get full access to this article
View all access options for this article.