基于激光雷达的城市交叉路口目标分类  

Object Classification of Urban Intersection Based on Lidar

作　　者：胡梦宽 顾晶 Hu Mengkuan;Gu Jing(College of Electronics&Information Engineering,Nanjing University of Information Science&Technology,Nanjing 210044,Jiangsu,China;College of Electronic Information Engineering,Wuri University,Wuri 214105,Jiangsu,China)

机构地区：[1]南京信息工程大学电子与信息工程学院,江苏南京210044 [2]无锡学院电子信息工程学院,江苏无锡214105

出　　处：《应用激光》2024年第11期141-148,共8页Applied Laser

基　　金：南京信息工程大学滨江学院车路协同雷达关键技术研究创新项目(2022r031)。

摘　　要：由于览沃系列激光雷达的非重复扫描特性,其即时扫描采样的运动目标会产生拖影形变、点云稀疏等问题。故提出了一种改善运动失真和点云密度的目标检测分类算法,该算法首先采用背景差分算法滤除背景点云,随后通过先验旋转平移矩阵融合5帧点云数据,同时采用可变阈值的快速欧几里得聚类算法对点云进行聚类;其次提取目标点云的几何信息、反射强度、方向梯度直方图等特征训练支持向量机分类器,实现对目标点云的分类;最后通过精确率和召回率等评价指标分析算法的分类性能。试验结果表明,该算法在三维激光雷达部署于城市交叉路口环境下对不同目标具有良好的分类和实时性能。The oscillating scanning pattern of Livox lidar leads to challenges such as drag deformation and sparse point clouds when scanning and sampling moving objects in real-time.This paper introduces an object detection and classification algorithm designed to mitigate motion distortion and enhance point cloud density.Firstly,the algorithm uses background extraction to separate the background points cloud,and then uses the priori rotation translation matrix to fuse the 5 frames point cloud data.At the same time,it uses the Fast Euclidean Clustering algorithm with dynamic threshold to cluster point clouds.Then the geometry information,intensity,histogram of oriented gradient and other features of the object point clouds are extracted to train the Support Vector Machine classifier to achieve the object classification.Finally,the classification performance of the algorithm is analyzed through evaluation indicators such as the precision and recall.The experimental results show that the algorithm has excellent classification and real-time performance for different objects under the condition that the lidar is deployed in the urban intersection environment.

关 键 词：目标分类 激光雷达 快速欧几里得聚类 支持向量机 

分 类 号：TN958.98[电子电信—信号与信息处理]