摘要: |
激光SLAM 通常通过多帧点云配准,完成位姿变换矩阵的估计,而点云中的动态点会降低
激光里程计的定位精度。为了减少动态点引入的误差,提出了一种动态点云识别算法,并结合该
算法改进了传统特征匹配策略,组成了动态环境下融合激光雷达和IMU 的激光里程计。通过约
束范围角与动态中心点,将点云快速分割成多个簇类,再借助IMU 信息,快速建立点云簇类配准
关系,从而去除动态点,最后根据簇类的对应关系进行约束,以提高匹配的精度与速度。使用
KITTI数据集和UGV 在多个情形下进行了实验。实验结果表明,该算法可以成功识别点云中的
多个动态对象,并通过去除动态点,有效地减少了位姿估计的累积误差。 |
关键词: 动态识别 激光里程计 点云分割 |
DOI: |
|
基金项目:中央高校基本科研业务费(2242021K1G008);东南大学国家自然科学基金剩余资金培育项目(9S20172204) |
|
Design of Lidar Odemeter Integrating Lidar and IMU in Dynamic Environment |
ZHANGTao,ZHANGChen,WEIHong-yu |
(schoolofInstrumentScienceandEngineering,SoutheastUniversity,Nanjing210096,China;KeyLabofMicroInertialInstrumentsandAdvancedNavigationTechnologyofMinistryof
Education,SoutheastUniversity,Nanjing210096,China) |
Abstract: |
Multi-framepointcloudregistrationisusuallyusedtoestimatetheposetransformation
matrix.Thedynamicpointsinthepointcloudreducethepositioningaccuracyofthelidarodometer.
Inordertoreducetheerrorintroducedbydynamicpoints,adynamicpointcloudrecognition
algorithmisproposed,andcombinedwiththisalgorithm,thetraditionalfeaturematchingstrategy
isimprovedtoformalaserodometerthatintegrateslidarandIMUinadynamicenvironment.The
pointcloudisquicklydividedintomultipleclustersbyrestrictingtherangeangleandthedynamic
centerpoint,andthenusingtheIMUinformation,theregistrationrelationshipofthepointcloud
clustersisquicklyestablished.Therebythedynamicpointsareremoved,andfinallytheprecision
andspeedofmatchingareimprovedbyconstrainingaccordingtothecorrespondingrelationshipof
theclusters.ExperimentsarecarriedoutinmultiplesituationsbyusingtheKITTIdatasetand
UGV.Theresultsshowthatthedesigneffectivelyreducethecumulativeerrorofposeestimation
byremovingdynamicpoints. |
Key words: Dynamicrecognition Lidarodometer Pointcloudsegmentation |