面向铁路无人机巡检的大范围点云配准方法
Large-scale Point Cloud Registration Method for Railway UAV Inspection
王志鹏 1邵长虹 2杨怀志 3秦勇 1薄一军 2古文超 2张丁慈 1耿毅轩1
作者信息
- 1. 北京交通大学先进轨道交通自主运行全国重点实验室,北京 100044
- 2. 京沪高速铁路股份有限公司,北京 100038
- 3. 京福铁路客运专线安徽有限责任公司,安徽合肥 230031
- 折叠
摘要
在我国铁路网络广阔、运输需求巨大的背景下,利用无人机配合高精度激光雷达进行铁路巡检,相较传统人工方法,不仅提高了效率,而且通过精准的点云数据配准与分析,构建出完善的三维铁路环境地图,为线路维护、故障排查及安全隐患定位提供可靠的数据支持.针对铁路及周边环境大规模非结构化数据的挑战,提出一种旋转不变性强、泛化能力优异的点云配准算法.通过在3DMatch公开数据集上训练模型,并在ETH数据集及京沪高铁数据集上进行测试,该算法展示了对未知数据集高效准确配准的能力,显著提升了无人机在铁路巡检中的应用价值,为铁路系统的安全运营与高效管理提供了强有力的技术支持.
Abstract
Under the background of China's vast railway network and huge transportation demand,the use of UAV in combination with high-precision laser radar for railway inspection not only improves efficiency compared with traditional manual methods,but also enables the development of a perfect 3D railway environment map through accurate point cloud data registration and analysis to provide reliable data support for line maintenance,troubleshooting and safety hazard positioning.In response to the challenge of large-scale unstructured data in railway and surrounding environment,a point cloud registration algorithm with strong rotation invariance and excellent generalization ability was proposed.By training the model on the 3DMatch public dataset and testing on the ETH dataset and the dataset of Beijing-Shanghai HSR,this algorithm has demonstrated its ability to efficiently and accurately register unknown datasets,significantly improving the application value of UAV in railway inspection and providing strong technical support for safe operation and efficient management of railway system.
关键词
京沪高铁/铁路巡检/无人机/点云配准/3DMatch数据集/RANSAC算法Key words
Beijing-Shanghai HSR/railway inspection/UAV/point cloud registration/3DMatch dataset/RANSAC algorithm引用本文复制引用
基金项目
中国国家铁路集团有限公司科技研发计划(P2022X001)
出版年
2024
NETL
NSTL
万方数据