Automatic extraction of high-voltage transmission pylons with multifeature constraints

Pylons are an important component of the transmission line, and its identification based on airborne Light Detection and Ranging (LiDAR) is crucial to power inspection. The efficient and high-precision extraction of pylon point clouds is important, especially in long-distance and large-scale applications, and is also conducive to massive data organization, parallel processing, and quantitative applications. The existing pylon extraction methods usually require a balanced and tremendous amount of training samples or lack sufficient terrain adaptability. Furthermore, these methods are vulnerable to tall objects, such as trees and buildings in the complex terrain environment of the mountainous areas. This study proposes an automatic pylon extraction method based on multifeature constraints. First, the height above the ground and the maximum vertical gap are designed on the basis of the spatial distribution of objects in the transmission corridor point clouds. Second, a series of preprocessing tasks, such as denoising and filtering, is performed on airborne LiDAR point clouds. Third, the pylon regions are quickly located on the basis of multifeature constraints, such as height difference and linearity, and the pylon center coordinates are calculated by using the layered density method and pylon structural symmetry. Finally, the point clouds of pylon regions are vertically sliced along the Z axis, and the nonpylon point clouds are eliminated layer by layer using the gap between the interference and the pylon vertical slicing. Airborne LiDAR point clouds in three different scenarios are utilized to evaluate the performance of the proposed method. The root mean square error of the pylon center coordinates are 0.04, 0.40, and 0.13 m. The precision, recall, and F1-value of the pylon extraction can reach up to 91.6%, 96.0%, and 93.5%. Compared with other pylon extraction methods, the qualitative analysis results show that the proposed method performs better in pylon area recognition, positioning error, and pylon point cloud extraction. Meanwhile, the proposed method successfully extracts pylons from variable terrain point clouds. Experimental results show that the proposed method can effectively extract pylons with high accuracy and strong terrain adaptability. In addition, the method does not need to train samples and consider class-imbalance problems. Furthermore, the proposed method can provide auxiliary information for postprocessing, such as scene classification and line hanging point extraction, and promote the application of airborne LiDAR for power inspection. © 2024 Science Press. All rights reserved.