| Title |
Development of Machine Learning-based Segmentation and Height Measurement Method for the Contact Wires |
| Authors |
정대현(Daehyeon Jeong) ; 이기원(Kiwon Lee) ; 박철민(Chulmin Park) ; 김동규(Dongkue Kim) |
| DOI |
https://doi.org/10.5370/KIEE.2024.73.2.382 |
| Keywords |
Contact wire; Height measurement; Trolley; Machine learning; Point cloud segmentation |
| Abstract |
Contact wires of catenary system are critical components of electric railways, providing electrical power to trains through contact with the pantograph. Improper installation of the wires can lead to abnormal wear on the contact strips of the pantograph, reducing performance and shortening its lifespan. Therefore, measurements of construction errors in the wires are frequently performed, currently conducted manually by on-site personnel. To reduce measurement time, labor requirements, and fatigue, this study proposes a method for automating the measurement of contact wire height by segmenting them from 3D point cloud data. Initially, wires are detected using Random Sample Consensus (RANSAC), and then Gaussian Mixture Models (GMMs) and Gaussian Mixture Regressors (GMRs) are used to refine the segmentation. A comparison of the measured height at 76 points using a laser rangefinder and the height from the algorithm yielded an average error of -1.2 mm, with a standard deviation of 3.4 mm. |