第60卷第2期/2023年1月/激光与光电子学进展0228005-1研究论文基于激光雷达的料堆特征提取方法优化张荠匀1，王建军1*，李旭辉1，王炯宇1，程霄霄1，王光彬21山东理工大学机械工程学院，山东淄博255049；2山东直通车科技有限公司，山东淄博255000摘要对料堆表面和形态进行特征提取是实现仓储自动化、智能化的前提与基础，为货料的自动存、取控制提供判断依据。为了提取料堆的形态与覆盖面特征，首先，采用激光雷达对料堆进行扫描，获取三维点云后使用融合算法进行预处理；其次，基于表面法向量的差异和空间距离差异对点云进行超体素聚类；最后，利用曲面凹凸判断方法对聚类后的三维点云曲面提取出凸面，从而实现了料堆表面形态的判断。实验结果表明，该方法可较好地识别料堆表面特征，识别误差小于3.11%，且不需要针对场景进行训练，可直接应用于不同料堆场景。关键词遥感与传感器；激光雷达；特征提取；超体素聚类；凹凸关系；区域生长中图分类号TN958.98文献标志码ADOI：10.3788/LOP212708OptimizationofFeature-ExtractionMethodforStockpiledMaterialsBasedonLiDARZhangJiyun1,WangJianjun1*,LiXuhui1,WangJiongyu1,ChengXiaoxiao1,WangGuangbin21SchoolofMechanicalEngineering,ShandongUniversityofTechnology,Zibo255049,Shandong,China;2ShandongThroughTrainTechnologyCo.,Ltd.,Zibo255000,Shandong,ChinaAbstractFeatureextractionofthesurfaceandformofstockpiledmaterialsisperformedforachievingtheautomationandintelligenceofwarehousing,anditprovidestheanalysisbasisfortheautomaticstorageandacquisitioncontrolofthematerials.First,thestockpiledmaterialisscannedbyusingLiDARtodetermineitsmorphologyandcoveragecharacteristics,a3Dpointcloudisobtained,andafusionalgorithmisusedtopreprocessthematerial.Second,thesupervoxelclusteringofpointcloudsisperformedbasedonthedifferenceofthesurfacenormalvectorandspatialdistance.Finally,theconvexsurfaceisextractedfromthe3Dpointcloudsurfaceafterclusteringbyusingtheconcaveandconvexjudgmentmethodtoanalyzethesurfaceshapeofthestockpiledmaterial.Theexperimentalresultsshowthatthemethodcanpreciselyrecognizethesurfacecharacteristicsofthestockpiledmaterial,andtherecognitionerrorislessthan3.11%.Theproposedmethodcanbedirectlyappliedtodifferentstock...
