北大中文核心期刊国外电子测量技术DOI:10.19652/j.cnki.femt.2204491基于路侧激光雷达的障碍物目标检测方法*杨建华赵轩郭全民方园园吴萍萍(西安工业大学电子信息工程学院西安710021)摘要:针对现有障碍物检测方法在复杂道路场景下存在地面分割欠精准、计算量大以及不同距离下的目标聚类困难问题,提出了一种基于路侧激光雷达的障碍物检测方法。在地平面分割方面,提出基于圆柱坐标系的改进扇形栅格模型以及最低点代表法优化种子点的选取,采用多地平面模型并通过随机采样一致性算法(RANSAC)实现地面拟合及分割。在障碍物聚类方面,构建KDTree加速聚类过程,提出划分区域及阈值自适应的方式改进欧氏聚类算法。实验结果表明,该方法在4种典型道路场景下对地面点的分割准确率均达到86%以上,且针对不同距离下的障碍物目标聚类准确率提升明显。关键词:路侧激光雷达;障碍物检测;地平面分割;目标聚类;栅格模型中图分类号:TN249;TN958.98文献标识码:A国家标准学科分类代码:510.20ObstacletargetdetectionmethodbasedonroadsideLiDARYangJianhuaZhaoXuanGuoQuanminFangYuanyuanWuPingping(ElectronicInformationEngineering,Xi'anTechnologicalUniversity,Xi'an710021,China)Abstract:Aimingattheproblemsoftheexistingobstacledetectionmethodsincomplexroadscenes,suchasinaccurategroundsegmentation,largeamountofcalculationanddifficulttargetclusteringatdifferentdistances,aroadsideLiDARbasedobstacledetectionmethodwasproposed.Intheaspectofgroundplanesegmentation,theimprovedfan-shapedgridmapbasedoncylindricalcoordinatesystemandthelowestpointrepresentationmethodareproposedtooptimizetheselectionofseedpoints.Thegroundfittingandsegmentationarerealizedbyusingmulti-planemodelandrandomsampleconsensusalgorithm(RANSAC).Intheaspectofobstacleclustering,KDTreeisconstructedtoacceleratetheclusteringprocess,andtheEuclideanclusteringalgorithmisimprovedbydividingtheregionandthresholdadaptive.Theexperimentalresultsshowthatthesegmentationaccuracyofthismethodforgroundpointsinfourtypicalroadscenesismorethan86%,andtheclusteringaccuracyofobstacletargetsatdifferentdistancesissignificantlyimproved.Keywords:roadsideLiDAR;obstacledetection;groundplanesegmentation;targetclustering;rastermodel收稿日期:2022-1...
