基于无人机传感器数据与词袋模型的点云自动化分类

激光点云分类是Li DAR数据应用的关键步骤和重要研究课题。针对Li DAR点云数据识别率低的问题,以体素化的点云为研究对象,提出了一种基于词袋模型的城区机载Li DAR点云数据分类方法。考虑到点云数据缺乏纹理信息,文中综合分析了点云数据和影像数据的特点,以点为单位提取描述点云的几何特征和影像特征分类特征;以体素为单位分割点云数据,并以体素为基础构建描述场景信息的词袋模型;最后基于随机分类器完成场景的分类。文中以ISPRS提供的Vaihingen数据作为实验数据。实验结果表明,本文提出的模型能有效地改善点云的分类质量,分类正确率能达到93%以上。

Point Cloud Auto-classification Based on UAV Sensor Data and Word Bag Model

Point cloud classification is a key step in LiDAR data application and an important research topic. Aiming at the problem of low recognition rate of LiDAR point cloud data, this paper presents a method of classification of airborne LiDAR point cloud data based on word bag model with voxel point cloud as the research object. Considering the lack of texture information in point cloud data, this paper analyzes the characteristics of point cloud data and image data synthetically, and extracts the geometric features and image feature classification characteristics of point cloud. Then the cloud data is divided in voxel and the phonetic model of the scene information is constructed based on the voxel. Finally, the classification of the scene is completed based on the random classification classifier. The Vaihingen data provided by ISPRS are used as experimental data. The experimental results show that the proposed model can be effectively improved the classification quality of point cloud, and the classification rate can get more than 93％.