地震后机载LiDAR点云的地物区分方法研究

The Ground-objects Classification Based on Post-earthquake Airborne LiDAR Data

利用机载激光雷达扫描(Light Detection and Ranging,LiDAR)技术所得点云进行震后倒塌建筑物提取时,树木与倒塌建筑物的点云特征十分相似,较难区分。为了快速准确获取震后房屋建筑物的受损情况,本文提出使用回波次数比特征指标,结合前人所提出的点云回波强度、归一化强度、最邻近点高差、法向量夹角、X向坡角和Y向坡角等特征的均值和标准差,利用K-最近邻分类法实现单体地物区分的方法。对2010年海地7.0地震震后机载LiDAR数据进行了地面点去除,分别选取了未倒塌建筑物、倒塌建筑物和树木各50个训练样本和各20个测试样本,计算了各因子的分布及其均值和标准差,在分析的基础上最终选取了可分性较强的8个分类特征,利用K-最近邻分类法对测试样本进行了分类,结果显示分类正确率可达85%以上。研究表明选取多个有效的LiDAR点云分类特征可以较好地区分震后未倒塌建筑物、倒塌建筑物和树木,提高震后建筑物震害程度判定的准确性,为应急救援及时提供较为准确的灾情信息支持。

It is difficult to distinguish the characteristics of the point cloud from the collapsed buildings, since the characteristics of the cloud and the collapsed building are similar. In this paper, we present the ratio of echo ratio and combined with the previous point cloud echo intensity, normalization intensity, altitude difference of the nearest neighbor, normal vector, X-slope and Y-slope of points, and compute the mean and standard deviation of every feature to obtain damage buildings after the earthquake. To discriminate single object, we use the classification algorithm based on K-Nearest Neighbor. We select 150 point clouds samples of 50 typical undamaged building, 50 collapsed building and 50 tree as samples from airborne LiDAR point cloud data which got after the 2010 Haiti earthquake with Mw 7.0 by the way of human-computer interaction and compute the mean and standard deviation of every feature. We apply K-Nearest Neighbor to classify test samples by 8 optimal factors chosen from means and standard deviations. The classification accuracy is 85%, which indicates that the optimal factors are effective and the proposed method in this study is capable of distinguishing between building （undamaged and damaged building） points and tree points, which can extract damaged buildings and support for emergency rescue after earthquake.