车载激光点云道路场景杆状地物分类研究  

Research on rod-shaped objects classification in road scene of vehicle-borne LiDAR point cloud

作　　者：臧静 李永强[1] 赵上斌 刘亚坤 杨亚伦 ZANG Jing;LI Yongqiang;ZHAO Shangbin;LIU Yakun;YANG Yalun(School of Surveying and Land Information Engineering,Henan Polytechnic University,Jiaozuo,Henan 454000,China)

机构地区：[1]河南理工大学测绘与国土信息工程学院,河南焦作454000

出　　处：《测绘科学》2022年第4期122-128,共7页Science of Surveying and Mapping

基　　金：国家自然科学基金项目(41771491)。

摘　　要：针对车载激光点云数据中杆状地物分类效果不理想以及单一分类算法具有局限性的问题,该文提出一种基于多重投票方式的改进引导聚集(Bagging)集成学习方法。根据地物点云特征值组成特征向量,利用样本集数据分别对多种机器学习算法进行训练并构建分类模型,获取每个分类器识别能力的先验知识;利用改进的Bagging集成分类算法对识别能力较强且可能存在互补信息的算法进行集成;采用多重投票方法实现杆状地物的自动分类。实验结果表明,该文算法对道路场景中杆状地物的分类精度可达98.58%,高于其他单分类器,对点云自动化分类具有一定的参考。Aiming at the problem that the rod-shaped objects classification of vehicle-borne laser point cloud data is not ideal and the single classification algorithm has limitations,an improved Bagging ensemble learning method based on multiple voting was proposed in this paper.The feature vector was composed according to the feature values of the feature points.The sample set data was used to train a variety of machine learning algorithms and build a classification model,which could obtain prior knowledge of the recognition ability of each classifier.The improved Bagging ensemble classification algorithm was used to integrate algorithms that have strong recognition capabilities and may have complementary information.The multiple voting method was used to achieve automatic classification of rod-shaped objects.Experimental results showed that the classification accuracy of the proposed algorithm was 98.58%,which was higher than other single classifiers,and it had reference for automatic point cloud classification.

关 键 词：点云分类 改进Bagging集成 杆状地物 决策树 支持向量机 

分 类 号：P231[天文地球—摄影测量与遥感]