基于改进DBSCAN聚类的激光雷达点云数据算法研究  

Research on LiDAR Point Cloud Data Algorithm Based on Improved DBSCAN Clustering

作　　者：孙浩然 杨家富[1] 王子洋 周梦飞 SUN Haoran;YANG Jiafu;WANG Ziyang;ZHOU Mengfei(College of Mehanical and Electronic Engineering,Nanjing Forestry University,Nanjing Jiangsu 210037,China)

机构地区：[1]南京林业大学机械电子工程学院,江苏南京210037

出　　处：《传感技术学报》2025年第2期301-308,共8页Chinese Journal of Sensors and Actuators

摘　　要：针对激光雷达采集到的原始三维点云数据量大导致无人车在障碍物检测过程中实时性差和准确率低的情况,提出了一种改进的密度噪声空间聚类(DBSCAN)算法。通过采用统计异常值去除(StatisticsOutlierRemoval)与体素网格(VoxelGrid)相结合的混合滤波算法对原始点云进行预处理,采用随机抽样一致性(RANSAC)算法对滤波后的点云进行地面分割以获取非地面点云,在传统DBSCAN算法中引入KD-tree,并将传统的欧氏距离聚类准则改进为标准化欧氏距离准则。改进后的DBSCAN算法在单帧点云数据的时效性方面得到提升,聚类效果准确可靠。Targeting at the poor real-time performance and low accuracy of unmanned vehicles in the obstacle detection process due to the large amount of original three-dimensional point cloud data collected by LiDAR,an improved density noise spatial clustering algo-rithm(DBSCAN)is proposed.The original point cloud is preprocessed by a hybrid filtering algorithm combining StatisticsOutlierRemoval and VoxelGrid.The filtered point cloud is segmented by using random sample consensus(RANSAC)algorithm to obtain non-ground point cloud.The KD-tree is introduced into the traditional DBSCAN algorithm,and the traditional Euclidean distance clustering criterion is improved to the standardized Euclidean distance criterion.The timeliness of the single frame point cloud data of the improved DBSCAN algorithm is improved,and the clustering effect is accurate and reliable.

关 键 词：无人车 激光雷达 点云处理 DBSCAN聚类 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]