惯性测量单元辅助的LiDAR动态点云剔除方法  

IMU-assisted LiDAR dynamic point cloud elimination method

作　　者：徐爱功[1] 高佳鑫 隋心[1] 袁庆[2] 陈志键 XU Aigong;GAO Jiarin;SUI Xin;YUAN Qing;CHEN Zhijian(School of Geomatics,Liaoning Technical University,Fuxin,Liaoning 123000,China;Engineering Survey Technology Application Research Institute,China Railway Siyuan Survey and Design Group Co.,Ltd.,Wuhan 430063,China)

机构地区：[1]辽宁工程技术大学测绘与地理科学学院,辽宁阜新123000 [2]中铁第四勘察设计院集团有限公司工程勘察研究院,武汉430063

出　　处：《测绘科学》2023年第5期173-182,共10页Science of Surveying and Mapping

基　　金：辽宁省重点研发计划项目(2020JH2/10100044);国家自然科学基金项目(42074012);辽宁省“兴辽英才计划”项目(XLYC2002101,XLYC2008034);辽宁省教育厅基础研究项目(LJ2020JCL016)。

摘　　要：针对动态场景下动态目标影响激光雷达同时定位与建图算法(LiDAR SLAM)的精度和成图效果问题,该文提出一种基于惯性测量单元(IMU)辅助的多层次模糊综合评价动态点云剔除方法。通过标定IMU/LiDAR外参统一两类传感器坐标系,再将每帧点云聚类分割为若干点云簇,以此为基础,利用IMU信息辅助建立相邻帧各点云簇间的配对关系,构建点云运动状态多层次模糊综合评价模型,判定各点云簇的运动状态,最终将动态点云簇从原始点云数据中剔除。为验证该文方法的可行性和精确性,设计了动态点云剔除实验,并将剔除动态点云后的点云数据输入激光雷达里程计与建图算法(LOAM)进行定位与建图。实验结果表明,该文方法动态点云的剔除成功率为98.67%,静态点云的误剔除率为2.01%,能够有效地提高点云数据质量。相比基于原始点云数据的LOAM算法,均方根误差降低了66.06%,最大误差降低了72.78%,实现了厘米级精度的定位,并且优化了建图效果。Aiming at the problem that dynamic targets affected the positioning accuracy and mapping effect of light detection and ranging simultaneous localization and mapping(LiDAR SLAM)in dynamic scenes,a multi-level fuzzy comprehensive evaluation dynamic point cloud elimination method based on the assistance of inertial measurement unit(IMU)was proposed.The coordinates of two sensors were unified by calibrating the external parameters of IMU/LiDAR.Then the point cloud of each frames was divided into several point cloud clusters.On this basis,the matching relationship between each point cloud cluster of adjacent frames was established by using IMU information,and the multi-level fuzzy comprehensive evaluation model of point cloud motion state was constructed to determine the motion state of each point cloud cluster.Finally,the dynamic point cloud clusters were removed from the original point cloud data.In order to verify the feasibility and accuracy of the method in this paper,the dynamic point cloud elimination experiment was designed and the point cloud data after eliminating the dynamic point cloud was input into lidar odometry and mapping(LOAM)for positioning and mapping.The experimental results showed that the success rate of eliminating dynamic point cloud was 98.67%,and the error rate of eliminating static point cloud was 2.01%,which could effectively improve the quality of point cloud data.Compared with the LOAM algorithm based on the original point cloud data,the root mean square error was reduced by 66.06%,and the maximum error was reduced by 72.78%.The centimeter level positioning accuracy was realized,and the mapping effect was optimized.

关 键 词：IMU LIDAR 动态点云 模糊综合评价模型 LOAM 

分 类 号：P237[天文地球—摄影测量与遥感]