第39卷第1期2024年1月Vol.39No.1Jan.2024液晶与显示ChineseJournalofLiquidCrystalsandDisplays基于统计局部特征描述与匹配的点云配准算法王鑫淼,李新春*,陶志勇(辽宁工程技术大学电子与信息工程学院,辽宁葫芦岛125100)摘要:针对ICP算法在初始位姿差、部分数据丢失和噪声干扰情况下鲁棒性差、配准精度低的问题,提出一种基于统计局部特征描述与匹配的点云配准算法。首先,用点云局部密度、点云拟合平面距离方差、高斯曲率和平均曲率构建一个四维的统计局部特征描述符,准确地描述查询点的局部特征;然后,通过点对间的特征差异进行对应点匹配,剔除错误点对,解决点云部分数据缺失和噪声干扰的问题;最后,使用平均匹配距离作为度量改进ICP算法,对点云进行配准,解决初始位姿较差时配准精度低的问题。实验结果表明,该算法在初始位姿差、部分数据丢失和噪声干扰情况下的配准精度提高至少1个量级,配准速率也有较大提升,在鲁棒性和配准精度方面均表现出明显优势。关键词:点云配准;特征描述;特征匹配;平均匹配距离;迭代最近点中图分类号:TP391.4文献标识码:Adoi:10.37188/CJLCD.2023-0053PointcloudregistrationalgorithmbasedonstatisticallocalfeaturedescriptionandmatchingWANGXinmiao,LIXinchun*,TAOZhiyong(SchoolofElectronicandInformationEngineering,LiaoningTechnicalUniversity,Huludao125100,China)Abstract:ApointcloudalignmentalgorithmbasedonstatisticallocalfeaturedescriptionandmatchingisproposedtoaddresstheproblemsofpoorrobustnessandlowalignmentaccuracyoftheICPalgorithminthepresenceofpoorinitialpositional,partialdataloss,andnoiseinterference.Firstly,a4-dimensionalstatisticallocalfeaturedescriptorisconstructedusingpointcloudlocaldensity,pointcloudfittingplanedistancevariance,Gaussiancurvature,andmeancurvaturetoaccuratelydescribethelocalfeaturesofthequerypoints.Then,thecorrespondingpointsarematchedbythefeaturedifferencebetweenpointpairstoeliminatethewrongpointpairsandsolvetheproblemsofmissingdataandnoiseinterferenceinpartofthepointcloud.Finally,themeanmatchingdistance(MMD)isusedasametrictoimprovethealignmentaccuracy.MMDisametrictoimprovetheICPalgorithmtoalignthepointcloudsandsolvetheproblemoflowalignmentaccuracywhenthein...