基于车载32线激光雷达点云的车辆目标识别算法  

Vehicle Target Identification Algorithm Based on Point Cloud of Vehicle 32-line Laser Lidar

作　　者：孔栋[1] 王晓原[1,2] 刘亚奇 陈晨 王方 

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

出　　处：《科学技术与工程》2018年第5期81-85,共5页Science Technology and Engineering

基　　金：汽车安全与节能国家重点实验室开放基金(KF16232); 国家自然科学基金(61074140,61573009,51508315,51608313); 山东省自然科学基金(ZR2014FM027,ZR2016EL19); 山东省社会科学规划研究项目(14CGLJ27); 山东省高等学校科技计划(J15LB07)资助

摘　　要：针对结构化道路环境中智能车识别周围360°范围内的车辆目标问题,由车载3D激光雷达采集的道路环境中车辆目标点云数据投影特征,提出识别车辆目标新算法。算法首先识别结构化道路边界,进而排除道路边界两旁障碍物的干扰和减少点云数据量;其次按雷达点云数据扫描和分布特征,利用改进K-means算法对道路区域内点云数据聚类。最后提取聚类目标内部特征点,并通过计算特征点构成向量的夹角或模的长度准确识别车辆目标。实验验证表明,该算法有效抑制了道路边界两旁障碍物的干扰,可以准确识别结构化道路区域内的车辆目标。Aiming at the vehicle target recognition problem in the 360° range around the intelligent vehicle,a new vehicle target identification algorithm is proposed based on the contour projection characteristics of obstacle target cloud obtained by 3D laser radar. Firstly,the structured road boundary is identified,thereby the interference of the obstructions on both sides of the road boundary is eliminated and the amount of cloud data is reduced; Secondly,based on the scanning and distribution characteristics of radar point cloud data,the improved K-means algorithm is used to cluster the point cloud data in the road area. Finally,the internal feature points of the clustering target are extracted,and the vehicle target is accurately identified by calculating the angle of the vector or the length of the vector. Experimental results show that the proposed algorithm can effectively suppress the disturbance of obstacle on both sides of the road boundary and can accurately identify the vehicle target in the structured road area.

关 键 词：交通安全 激光雷达 改进K-means聚类 车辆目标识别 

分 类 号：TP242.6[自动化与计算机技术—检测技术与自动化装置]