第11卷第2期导航定位学报Vol.11,No.22023年4月JournalofNavigationandPositioningApr.,2023引文格式:崔善尧,贾述斌,黄劲松.图像与激光点云融合的实时目标定位算法[J].导航定位学报,2023,11(2):99-105.(CUIShanyao,JIAShubin,HUANGJingsong.Real-timetargetpositioningalgorithmbasedonimageandlaserpointcloudfusion[J].JournalofNavigationandPositioning,2023,11(2):99-105.)DOI:10.16547/j.cnki.10-1096.20230211.图像与激光点云融合的实时目标定位算法崔善尧,贾述斌,黄劲松(武汉大学测绘学院,武汉430079)摘要:针对单一传感器在自动驾驶实时目标定位中的局限性,提出一种基于图像与激光点云融合的实时目标定位算法:对图像进行目标检测,得到目标在图像坐标系中的区域;然后将激光点云投影到图像坐标系中,使用基于密度的带有噪声的空间聚类(DBSCAN)算法对目标区域内的少量点云进行聚类以提取目标真实点云,降低计算量的同时获取目标准确的位置信息。实验结果表明,该算法可以有效提升在遮挡情况下目标定位的准确性;并通过在嵌入式平台上进行部署,验证了该算法在算力受限的平台上也具有较好的实时性。关键词:自动驾驶;图像检测;点云聚类;数据融合;实时目标定位中图分类号:P228文献标志码:A文章编号:2095-4999(2023)02-0099-07Real-timetargetpositioningalgorithmbasedonimageandlaserpointcloudfusionCUIShanyao,JIAShubin,HUANGJingsong(SchoolofGeodesyandGeomatics,WuhanUniversity,Wuhan430079,China)Abstract:Aimingatthelimitationsofthesinglesensorinreal-timetargetlocalizationinautonomousdriving,thepaperproposedareal-timetargetlocalizationalgorithmbasedonfusionofimagesandlaserpointclouds:thetargetdetectionontheimagewasperformedtoobtaintheareaofthetargetintheimagecoordinatesystem;andthedensity-basedspatialclusteringofapplicationswithnoise(DBSCAN)algorithmwasusedtoclusterasmallnumberofpointcloudsinthetargetareatoextracttherealtargetpoints;thentheaccuratelocationofthetargetwasobtainedwhilereducingtheamountofcomputation.Experimentalresultsshowedthattheproposedalgorithmcouldeffectivelyimprovetheaccuracyoftargetpositioninginthecaseofocclusion;andthroughdeployingitontheembeddedplatform,itcouldbeverifiedthatthealgorithmwouldhavegoodreal-timeperfo...