基于点云特征的改进RANSAC地面分割算法  

Improved RANSAC ground segmentation algorithm based on point cloud features

作　　者：隋心[1] 王思语 罗力 陈志键 史政旭 张杰 郝玉婷 SUI Xin;WANG Siyu;LUO Li;CHEN Zhijian;SHI Zhengxu;ZHANG Jie;HAO Yuting(School of Geomatics,Liaoning Technical University,Fuxin,Liaoning 123000,China;Liaoning Geospatial Results Application Center,Liaoning Natural Resources Affairs Service Center,Shenyang 110086,China)

机构地区：[1]辽宁工程技术大学测绘与地理科学学院,辽宁阜新123000 [2]辽宁省自然资源事务服务中心辽宁省地理空间成果应用中心,沈阳110086

出　　处：《导航定位学报》2024年第1期106-114,共9页Journal of Navigation and Positioning

摘　　要：针对室外复杂场景下,轻量级和地面优化的激光雷达里程计与测图(LeGO-LOAM)算法由于地面分割不精确而导致算法定位精度降低的问题,提出一种基于改进随机一致性采样(RANSAC)的多线程地面分割算法:相较于传统RANSAC算法,该算法舍弃从全部原始数据中随机选取种子点拟合地面模型的迭代方式,首先利用点云高程、曲率等点特征信息挑选出所有小于高程、曲率等阈值的种子点以构建种子点集合,并根据种子点集合中的种子点数量判断是否需要多线程处理;然后根据判断结果从种子点集合中选择种子点子集进行地面拟合;最后比较各地面模型所包含的点云数量以获得最优地面模型参数以及地面点云集;地面分割精度的提高有效地降低了LeGO-LOAM算法的定位误差。实验结果表明,在室外复杂场景下所提出的地面分割算法分割效果更好,杂点更少;相较于原LeGO-LOAM算法,改进算法的定位误差降低至3.73 m,平面均方根误差降低了20.8%。Aiming at the problem of reduced positioning accuracy of lightweight and ground-optimized light detection and ranging odometry and mapping(LeGO-LOAM)algorithm caused by inaccurate ground segmentation in outdoor complex scenes,the paper proposed a multi-threaded ground segmentation algorithm based on improved random sample consensus(RANSAC):compared with the traditional RANSAC algorithm,the iterative method of randomly selecting seed points from all the raw data was abandoned to fit the ground model,but the point cloud elevation,curvature and other point feature information were used to select all seed points that are less than the elevation,curvature and other thresholds to construct a seed point set,and whether multi-threaded processing was needed was determined based on the number of seed points in the seed point set;then,based on the judgment results,a subset of seed points from the seed point set was seletct for ground fitting;finally,the number of point clouds contained in each ground model was compared to obtain the optimal ground model parameters and ground point cloud set;the improvement of ground segmentation accuracy could effectively reduce the positioning error of the LeGO-LOAM algorithm.Experimental results showed that the proposed ground segmentation algorithm would perform better in outdoor complex scenes,with fewer noise points;moreover,compared with the original LeGO-LOAM algorithm,the improved algorithm could reduce the positioning error to 3.73 m and decrease the root mean square error in the plane by 20.8%.

关 键 词：轻量级和地面优化的激光雷达里程计与测图(LeGO-LOAM) 随机一致性采样(RANSAC) 地面分割 室外定位 

分 类 号：P228[天文地球—大地测量学与测量工程]