Abstract: The research on non-contact detection of catenary of electrified railway is of great significance to ensure the safe operation of the railway. The detection work requires a large number of accurate contact point cloud data support. At present, there is a problem that it is difficult to provide accurate contact point cloud data support due to the difficulty of segmentation between catenary components. Aiming at this problem, this paper proposes a catenary extraction method based on the combination of multi-level index and moving vector. Firstly, the railway scene data is simplified by multi-level index frame.Then the center point set at the bottom of the pillar is obtained by constructing the extraction channel through the trajectory line to calculate the moving vector along the rail. Finally, the attitude of the secondary index frame is adjusted to achieve the accurate extraction of the catenary. In this paper, parameter analysis and comparison experiments are designed, and experimental analysis is carried out in the 10 km railway scene. The results show that the precision, recall and F1 of the algorithm in this paper are about 99%, which are better than the reference algorithm, so the algorithm in this paper can adapt to complex scenes.

Key words: catenary extraction, multi-level index, 3D LiDAR, point cloud neighborhood search, movement vector