| 禹鑫燚,何伟琪,崔朱帆,欧林林.基于虚幻引擎与三维点云的列车障碍物检测仿真系统设计[J].高技术通讯(中文),2023,33(10):1077~1089 |
| 基于虚幻引擎与三维点云的列车障碍物检测仿真系统设计 |
| Design of railway obstacle detection simulation system based on unreal engine and 3D point cloud |
| |
| DOI：10. 3772/ j. issn. 1002-0470. 2023. 10. 008 |
| 中文关键词: 虚幻引擎； 列车仿真环境； 三维点云； 障碍物检测； 神经网络 |
| 英文关键词: unreal engine, railway simulation, 3D point cloud, obstacle detection, neural network |
| 基金项目: |
| 作者 | 单位 | | 禹鑫燚 | （浙江工业大学信息工程学院杭州 310023） | | 何伟琪 | | | 崔朱帆 | | | 欧林林 | |
|
| 摘要点击次数: 688 |
| 全文下载次数: 576 |
| 中文摘要: |
| 随着轨道技术的高速发展，列车安全的主动防护技术在列车行驶的过程中极为重要，例如非接触式障碍物检测系统。然而在真实列车运行的铁路环境中难以采集到带有障碍物的场景，并存在采集实验数据效率低的问题，如图像和点云数据。本文设计了一种具有虚拟激光雷达的轨道列车障碍物检测仿真系统，为轨道列车非接触式障碍物检测的研究提供可靠的实验环境和点云采集传感器。为了精确识别列车前方轨道位置，提出一种基于体素分割和稀疏卷积算子构建点云语义分割网络识别轨道点云的方法。基于最小二乘法拟合离散的轨道点云获得连续的安全行车区域，并且基于平面分割和点云聚类确定了障碍物的三维位置。通过采集仿真环境中3种轨道场景的点云，制作数据集用于训练网络模型，并通过采集带有障碍物场景的点云测试所提出的障碍物检测方法。实验结果验证了仿真系统可行性以及障碍物检测算法的有效性。 |
| 英文摘要: |
| With the development of railway technology, active protection technology for train safety is extremely important in train running, such as a non-contact obstacle detection system. However, obstacle scenarios are difficult to be captured in a real train-running railroad environment. There exists some difficulty in collecting experimental data, such as image and point cloud data. In this paper, a railway simulation system and a virtual lidar are designed to provide a reliable test environment and point cloud acquisition sensors for railway non-contact obstacle detection research. A point cloud semantic segmentation network is proposed to accurately identify the rail located in front of the train based on voxel partition and sparse convolution. A continuous safe driving area is obtained by fitting discrete rail point clouds based on the least squares algorithm. A method for calculating the 3D position of obstacles is presented based on plane segmentation and point cloud clustering. By collecting the point clouds of three railway scenes in the simulation environment, the datasets are made for training the network model. Furthermore, point clouds with obstacle scenes are collected to test the obstacle detection method. The experimental results verify the feasibility of the simulation system and the effectiveness of the proposed obstacle detection method. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
