车载雷达点云的结构化道路边界提取方法  

Structured Road Boundary Extraction Algorithm of Vehicle Radar Point Cloud

作　　者：孔栋[1] 王晓原[1,2] 孙亮[1] 王方[1] 陈晨 

机构地区：[1]山东理工大学交通与车辆工程学院,山东淄博255049 [2]清华大学汽车安全与节能国家重点实验室,北京100084

出　　处：《河南科技大学学报（自然科学版）》2018年第4期57-62,共6页Journal of Henan University of Science And Technology:Natural Science

基　　金：国家自然科学基金项目(61074140;61573009);山东省自然科学基金项目(ZR2014FM027)

摘　　要：为了提高结构化道路边界检测的准确性与鲁棒性,结合非参数变点统计方法,提出了一种基于32线激光雷达三维点云的道路边界提取算法。基于结构化道路区域和非道路区域存在一定高程跳变特征,该算法利用非参数变点统计,对激光雷达扫描的道路环境三维点云数据中突变的z坐标值进行标记,并提取对应的候选道路边界点(x,y)。利用道路边界方向的最大期望(EM)聚类算法,对候选道路边界点进行聚类去噪。利用最小二乘法拟合道路边界,在不同光照条件下的校园结构化直、弯道路环境进行实车实验,统计直道1 030帧数据和弯道650帧数据。仿真结果表明:算法识别准确性较高且检测距离达18 m,耗时约28 ms,可满足智能车实时性要求。In order to improve the accuracy and robustness of structured road boundary detection,the road boundary extraction algorithm based on 32-line lidar 3 D point cloud was proposed in combination with nonparametric change point statistical method. Based on the existence of certain elevation jumps in structured road and non-road areas,non-parametric change-point statistics were used to mark the z-coordinate values of sudden changes in 3 D point cloud data collected by lidar. The corresponding candidate road boundary points( x,y)were extracted. By using the maximum expectation( EM) clustering algorithm of road boundary direction,the extracted candidate road boundary points were clustered and denoised. The least squares method was used to fit the road boundary. Real car experiments were carried out in different light conditions of the campus structure straight and curved road environment. The data of 1 030 frames of straight road and 650 fromes of curved road were obtained. The simulation results show that the recognition of the algorithm has high accuracy with detection distance of 18 m and about 28 ms,which can meet the real-time requirement of intelligent vehicle.

关 键 词：智能车 激光雷达点云 非参数变点统计 道路边界提取 

分 类 号：TP242.6[自动化与计算机技术—检测技术与自动化装置]