面向磁浮轨道的多源点云数据的混合滤波方法  

Hybrid Filtering Method for Multisource Point Cloud Data of Maglev Tracks

作　　者：张玉鑫 张雷 欧冬秀[1,3] ZHANG Yuxin;ZHANG Lei;OU Dongxiu(School of Traffic and Transportation Engineering,Tongji University,Shanghai 201804,China;Shanghai Multi-Network and Multi-Mode Rail Transit Collaborative Innovation Center,Tongji University,Shanghai 201210,China;Shanghai Key Laboratory of Structural Durability and System Safety of Rail Transit,Tongji University,Shanghai 201804,China)

机构地区：[1]同济大学交通运输工程学院,上海201804 [2]同济大学上海市多网多模式轨道交通协同创新中心,上海201210 [3]同济大学上海市轨道交通结构耐久与系统安全重点实验室,上海201804

出　　处：《计算机工程》2024年第9期54-62,共9页Computer Engineering

基　　金：国家自然科学基金面上项目(52172329)。

摘　　要：在磁浮轨道的仿真数据处理过程中,磁浮轨道点云数据的滤波提取是重要环节之一,实际应用应根据待提取的磁浮数据特性,采用高效的滤波方法。磁浮轨道的点云数据对象主要包括由无人机(UAV)倾斜摄影获取的磁浮轨道的图像数据并经过三维重建后形成的稠密点云数据、由手持式激光雷达扫描磁浮轨道获取的激光点云数据。根据这两种点云的数据特性,考虑磁浮轨道四周复杂场景的点云环境,分别对两种点云进行混合滤波。首先,对激光点云数据采用八叉树下采样方法,有效降低了点云数据的数量级,节省了运行时间。然后,分别对激光点云与稠密点云数据采用布料模拟滤波(CSF)方法,过滤了地平面点云数据,保留了非地面点云数据;采用统计离群点去除(SOR)滤波方法,筛除了大量离群点;根据磁浮轨道特征,采用直通滤波过滤了坐标范围外的点云数据。实验结果表明,在不影响磁浮轨道结构的前提下,对于采用八叉树下采样方法的激光点云数据和没有采用八叉树下采样的稠密点云数据,该方法的滤波率分别为86.15%和64.76%,经混合滤波后的两种点云数据的结构近似,点云数量处于同一数量级,为磁浮轨道点云特征提取等后续任务提供了有效保障。In the simulation data processing of maglev tracks,the filtering and extraction of maglev track point cloud data is an important link.Thus,practical applications should adopt an efficient filtering method according to the characteristics of the maglev data to be extracted.The point cloud data objects of the maglev track primarily include the image data of the maglev track,which is obtained by Unmanned Aerial Vehicle(UAV)oblique photography and formed into dense point cloud data after 3D reconstruction,and the laser point cloud data,which is obtained by handheld lidar scanning of the maglev track.Based on the data characteristics of these point clouds and considering the complex scenes around the maglev track,the two types of point clouds are mixed and filtered.First,the octree downsampling method is used for laser point cloud data,which effectively reduces the order of magnitude of the point cloud data and saves running time.The Cloth Simulation Filtering(CSF)method is then used on the laser point cloud and dense point cloud data to filter the ground plane point cloud and retain the non-ground point cloud data,respectively.A Statistical Outlier Removal(SOR)filtering method is used to screen a large number of outliers.Based on the characteristics of the maglev track,point clouds outside the coordinate range are filtered through straight-through filtering.On the premise of not changing the structure of the maglev track,the experimental results show that the filtering rates of the proposed method are 86.15%and 64.76%for the octree-downsampled laser point cloud data and the dense point cloud data without octree downsampling,respectively.These two point cloud datasets have similar structural ranges after hybrid filtering and a number of point clouds of the same order of magnitude,which can be effective for methods such as feature extraction of point clouds in maglev orbits.

关 键 词：磁浮轨道 多源点云数据 八叉树下采样 布料模拟滤波 统计离群点去除滤波 

分 类 号：U213.2[交通运输工程—道路与铁道工程]