第60卷第2期/2023年1月/激光与光电子学进展0228002-1研究论文使用冯·米塞斯分布提取特征的点云精简方法刘源，左小清＊，李勇发，杨栩，周定义，黄琨昆明理工大学国土资源工程学院，云南昆明650093摘要针对点云精简算法提取特征时较为依赖传统参数、特征提取不全面和特征边界易丢失等问题，提出一种使用冯·米塞斯（vMF）分布提取特征的点云精简方法。该方法首先利用邻域重心点构建向量，通过与法向之间的夹角关系设置阈值划分曲面，降低噪声对细节特征的影响；然后利用冯·米塞斯分布提取曲面点的优先度，实现全局特征提取；最后基于特征进行八叉树分级精简。实验结果表明：所提方法可有效提取细节特征，相比于基于曲率、Hausdorff距离的方法所提取到的特征，有着更好的特征提取效果；利用基于曲率、栅格、随机的精简算法与所提方法进行关于重建结果、3D偏差、定量分析的对比，证明所提精简方法效果更优。所提方法为点云特征提取和精简提供一种新的思路。关键词点云精简；特征提取；冯·米塞斯分布；八叉树中图分类号TP391文献标志码ADOI：10.3788/LOP212641PointCloudSimplificationMethodUsingvonMises-FisherDistributiontoExtractFeaturesLiuYuan,ZuoXiaoqing＊,LiYongfa,YangXu,ZhouDingyi,HuangKunFacultyofLandResourcesEngineering,KunmingUniversityofScienceandTechnology,Kunming650093,Yunnan,ChinaAbstractAddressingtheissuesofpointcloudsimplificationalgorithmsthatrelyontraditionalparameterswhenextractingfeatures,whichisnotcomprehensiveandeasytolosefeatureboundaries,thisstudyprovidesapointcloudsimplificationapproachusingvonMises-Fisher(vMF)distributiontoextractfeatures.Thismethodfirstusesaneighborhoodcenterpointtocreateavector,dividesthesurfacethroughthethresholdoftherelationshipwiththenormaldirection,reducestheimpactofnoiseonthefinerfeatures.Then,thepriorityofsurfacepointsisextractedbyusingvMFdistributiontorealizeglobalfeatureextraction.Finally,octreehierarchicalsimplificationisoperatedbasedonfeatures.Experimentsdescribedthatthemethodinthisstudycansuccessfullyextractdetailedfeatures.ComparedwithmethodsbasedoncurvatureandHausdorffdistance,ithasabetterfeatureextractioneffect.Thesimplificationalgorithmsbasedoncurvature,grid,andrandom,andtheproposedmethod...
