Accurate LiDAR-Based Semantic Classification for Powerline Inspection

Mandatory power grid inspection includes the periodic measurement of the distances between vegetation and electrical assets. This paper presents an efficient and accurate method for segmenting LiDAR-based maps leveraging both LiDAR reflectivity and point cloud geometric information to classify points into Powerline, Tower, Vegetation, and Soil. It is based on two steps: 1) initial online segmentation using the LiDAR reflectivity and local point cloud spatial distribution and 2) off-line segmentation refinement using the cloud global spatial distribution. The method obtains a very accurate classification and enables accurate measurement of distances from vegetation to the electrical assets. Its performance was evaluated in sets of powerline inspection experiments in two power grid scenarios with different conditions and vegetation.