文章编号：1671-7872(2023)02-0198-07基于方向直方图签名描述符的点云配准方法赵卫东，程星，陈香梅(安徽工业大学电气与信息工程学院，安徽马鞍山243032)摘要：针对复杂场景下传统点云配准精度与效率低、鲁棒性差，难以配合机器人进行工业作业，提出一种基于方向直方图签名(SHOT)描述符的点云配准方法。对模型点云和场景点云采用体素重心降采样预处理，对降采样后的点云采用内部形状签名(ISS)提取特征点；计算特征点SHOT描述符并构建KD树快速检索特征相似的点对；采用随机采样一致(RANSAC)去除误匹配点对并完成粗配准，获取点云粗配准初始位姿，联合迭代最近点(ICP)完成精配准。实验结果表明：复杂场景下，本文方法能够快速识别定位H型钢和热电偶，配准用时2s，配准精度在5mm以内；与传统ICP算法相比，本文方法配准精度更高、鲁棒性更好，能够对一定程度遮挡、残缺物体点云进行识别，满足工业要求。关键词：复杂场景；点云配准；方向直方图签名描述符；机器人中图分类号：TP242.6文献标志码：Adoi：10.12415/j.issn.1671−7872.22255PointCloudRegistrationMethodBasedonSignatureofHistogramofOrientationDescriptorZHAOWeidong,CHENGXing,CHENXiangmei(SchoolofElectrical&InformationEngineering,AnhuiUniversityofTechnology,Maanshan243032,China)Abstract：Inviewofthelowaccuracy,efficiencyandrobustnessoftraditionalpointcloudregistrationincomplexscenes,whicharedifficulttocooperatewithrobotsinindustrialoperations,apointcloudregistrationmethodbasedonsignatureofhistogramsoforientations(SHOT)descriptorwasproposed.Themodelpointcloudandfieldpointcloudwerepre-processedbydownsamplingthecenterofgravityofvoxels,andtheinternalshapesignature(ISS)wasusedtoextractfeaturepointsfromthedownsampledpointcloud.TheSHOTdescriptorsoffeaturepointswerecalculated,andtheKDtreewasconstructedtoquicklyretrievethepairsofpointswithsimilarfeatures.Therandomsamplingconsistency(RANSAC)algorithmwasusedtoremovethemis-matchedpointpairsandcompletethecoarsealignment,obtaintheinitialpositionofthecoarsealignmentofthepointcloud,andcompletethefinealignmentbycombiningwithiterativeclosestpoint(ICP).TheexperimentalresultsshowthatthismethodcanquicklyidentifyandlocateH-beamsandthermocouplesincomplexscene...
