Document Type : Original Article
Department of Photogrammetry and Remote Sensing, School of Surveying and Geospatial Eng., University of Tehran, Tehran, Iran
Department of Communications, School of Electrical and Computer Eng., Faculty of Eng., University of Tehran, Tehran, Iran
Dept. of Geography and Environmental Management, University of Waterloo, Waterloo, Ontario, Canada
Mobile Laser Scanning (MLS) systems have been used for power line inspection in a fast and precise fashion. However, manually processing of huge LiDAR point clouds is tedious and time-consuming. Thus, an automated method is needed. This study proposes a machine learning-based method for automated detection of power lines from MLS point clouds. The proposed method consists of three main steps: pre-processing, line extraction using Support Vector Machine (SVM), and post-extraction. In the pre-processing step, noisy and low-height points are eliminated after sectioning the collected point clouds. This step considerably reduces the volume of point clouds by 90%. Then, the point features including linearity, planarity, verticality, and the largest component of Principal Component Analysis (PCA) are used as the best-fitted descriptors for power line detection. After training the SVM by a small section of points, SVM properly classified the point clouds with about 97% and 98% accuracies regarding precision and recall, respectively. In the final step, a post-extraction is required to eliminate false points in the power line class. This step improved the recall from 98% to 99.4% and decreased slightly the precision accuracy from 97% to 95.5%. The results demonstrated that the proposed method works rapidly, about 14 seconds per section with an average of 5 million points in each section.