一种基于欧氏簇提取的城市点云分类方法

A City Point Cloud Classification Method Based on Euclidean Extract

摘要: 点云分类作为数字城市建设的重要一步，其准确定和完整性将直接影响后续建模的精度。本文从点云空间结构出发，采用一种以簇为单位的点云分类思想，希望能在保证地物完整性的同时，也能达到一种良好的分类效果。首先是采用一种渐进形态学的方式对点云进行滤波处理，除去地面点使地物具有更好的可分性，再此基础上利用欧氏距离进行点云聚类；计算簇中所有点在不同尺度下的特征值，利用树木和建筑两个样本训练分类器；最后通过阈值分割提取建筑物簇。从结果上看，该方法基本能够达到建筑物的完整分类。

Abstract: As an important step in the construction of digital city, the accuracy and completeness of point cloud classification will directly affect the accuracy of subsequent modeling. Based on the spatial structure of point cloud, this paper adopts a point cloud classification idea with cluster as unit, hoping to achieve a good classification effect while ensuring the integrity of ground objects. Firstly, a progressive morphological method is used to filter the point cloud, and the ground points are removed to make the ground objects more separable. On this basis, the Euclidean distance is used to cluster the point cloud; the eigenvalues of all points in the cluster at different scales are calcu-lated, and the classifier is trained using two samples of trees and buildings; finally, building clusters are extracted by threshold segmentation. From the results, the method can basically achieve the complete classification of buildings.