An automatic framework for pylon detection by a hierarchical coarse-to-fine segmentation of powerline corridors from UAV LiDAR point clouds

Abstract

This paper proposes an automatic framework for pylon detection by a hierarchical coarse-to-fine segmentation of powerline corridors from UAV laser scanning point clouds. To this end, the proposed framework starts by roughly detecting the pylon location using the voxel-based height features derived from powerline corridor distribution in the vertical direction. The roughly detected pylons are then fed into the fine-grained pylon segmentation step, from which the fine-grained pylon points are learned by leveraging the shape prior knowledge. The idea behind the fine-grained is that most of the pylons can be cut horizontally into a series of rectangular cross-sections whose sizes from top to bottom are growing at a constant rate. By this linear growth relationship, the distorted cross-sections, which most commonly occur at pylon legs due to the influence of the attachments, such as trees and brush, can be accurately restored using the linear least squares regression. The performance of the proposed method was evaluated on two datasets over hilly and flat landforms. Our evaluation results showed that for powerlines in flat terrain, the proposed method achieved a precision of 99.8%, recall of 99.5%, and F1-score of 99.7%. On hilly terrain, a slightly lower performance was obtained, with a precision of 98.8%, recall of 97.8%, and F1-score of 98.3%. The proposed method’s accuracy is on par with or even better than other mainstream pylon detection algorithms.

Keywords

• Powerline corridor monitoring
• Pylon detection
• Hierarchical segmentation
• Shape prior knowledge
• Linear least squares regression
• MBR algorithm