Research on the Rapid Extraction Method of Overhead Line Laser Point Cloud Based on Fusion Segmentation

Unmanned aerial vehicles equipped with LiDAR (Light Detection and Ranging) systems can quickly acquire 3D point cloud data of transmission lines and have become a new technical means for intelligent grid inspection. However, the large amount of point cloud data collected by LiDAR and the complexity of the point cloud scene limit the accuracy and speed of the overhead conductor point cloud extraction algorithm. To address this issue, this paper proposes a rapid extraction method for 3D point clouds of overhead lines that integrates visible light structural features. Firstly, an improved DeepLabv3+ network is used to segment the visible light images of overhead lines, successfully locating the overhead line area from a complex background. Secondly, the MCMLSD line detection method is introduced to fill in the completeness of the overhead line area after DeepLabv3+ segmentation, solving the problem of discontinuity in the DeepLabv3+ overhead line segmentation area. Finally, based on the target area of overhead lines in visible light images, the rapid extraction of 3D point clouds of overhead lines in full-scene point cloud data is completed. Experiments have shown that this method has good robustness in scenes such as sky, land, and forests, providing multi-modal information for unmanned aerial vehicle automatic inspection of overhead lines and promoting the development of this field.