基于车载LiDAR点云数据的杆状地物分类模型  

Pole-shaped surface feature classification model based on vehicle-mounted LiDAR point cloud data

作　　者：金帅 蒋飞熊 陈进华 JIN Shuai;JIANG Feixiong;CHEN Jinhua(Zhejiang Provincial Land Survey and Planning Company Limited,Hangzhou,Zhejiang 310030,China;Zhejiang Academy of Surveying and Mapping,Hangzhou,Zhejiang 310030,China)

机构地区：[1]浙江省国土勘测规划有限公司,浙江杭州310030 [2]浙江省测绘科学技术研究院,浙江杭州310030

出　　处：《北京测绘》2024年第12期1718-1723,共6页Beijing Surveying and Mapping

基　　金：国家自然科学基金(42261074)。

摘　　要：杆状地物是道路场景中的重要基础设施,研究如何对其开展自动化识别与分类尤为重要,可为城市数据更新、智慧城市提供支持。本文选取车载激光雷达(LiDAR)点云数据作为研究对象,提出一种优化的引导聚集(Bagging)集成学习技术,旨在解决当前分类算法所面临的局限性挑战。首先,对原始点云进行滤波处理,得到非地面点;其次,获取杆状地物点云特征值并组成特征向量,同时构建机器学习分类模型;最后,使用单一分类器与本文提出的改进Bagging集成学习方法开展杆状地物分类实验。实验结果表明,本文提出的基于多重投票方式的改进Bagging集成学习方法能够有效实现杆状地物的提取与分类,精度可达98.47%。本文的研究可为基于点云数据的地物分类提供一定借鉴与参考。Pole-shaped surface features are important infrastructure in road scenes,and it is particularly important to study automated recognition and classification of these surface features,which can provide support for urban data updates and smart cities.This paper selected vehicle-mounted light detection and ranging(LiDAR)point cloud data as the research object and proposed an optimized bootstrap aggregating(Bagging)ensemble learning technique to address the limitations and challenges faced by current classification algorithms.Firstly,the paper filtered the original point cloud to obtain non-ground points.Secondly,the paper obtained the feature values of the pole-shaped surface feature point clouds and formed a feature vector,while constructing a machine learning classification model.Finally,experiments were conducted on pole-shaped surface feature classification by using a single classifier and the improved Bagging ensemble learning method.The experimental results show that the improved Bagging ensemble learning method based on multiple voting methods in this paper can effectively achieve the extraction and classification of pole-shaped surface features,with an accuracy of 98.47%.The research in this paper can provide certain references and guidance for point cloud data-based surface feature classification.

关 键 词：杆状地物 车载激光雷达点云 支持向量机(SVM) 改进引导聚集(Bagging)集成 分类提取 

分 类 号：P237[天文地球—摄影测量与遥感]