首页|基于无人机和点云滤波的边坡灾害识别方法

基于无人机和点云滤波的边坡灾害识别方法

扫码查看
原文链接
  • NETL
  • NSTL
  • 万方数据
为解决传统公路边坡灾害识别方法风险大、成本高以及植被覆盖情况下灾害识别困难的问题,采用无人机倾斜摄影与机器学习技术,构建了一种面向复杂地形环境的点云滤波和灾害识别方法.首先通过无人机倾斜摄影技术获取多角度高清边坡影像,重构高精度边坡三维点云模型;接着利用支持向量机(SVM)机器学习算法,训练和建立基于公路边坡表面形态特征的SVM点云分类模型,对边坡地面与植被点云进行识别与分类,获取滤除植被的边坡表面点云数字高程模型(DEM);最后采用DEM差分算法(DoD)对不同调查时期获取的边坡表面DEM进行地形变化检测,根据三维模型变化检测结果获取边坡灾害具体信息以及变化云图,从而实现公路边坡隆起、塌陷、落石等灾害的自动识别.将该技术应用于长沙市某公路边坡工程灾害调查,成功识别出的边坡隆起区域面积约为621.93 m2,平均隆起高度 1.13 m,塌陷区域面积约为 460.42 m2,平均塌陷高度 0.82 m,边坡灾害区域的总土方量约为 1 081.06 m3,构建的植被点云滤波方法准确率优于 98.4%.研究结果表明:构建的基于SVM与DoD算法的点云滤波和无人机边坡灾害自动识别方法,具备准确滤除植被、快速识别及量化边坡灾害的能力,且对复杂地形条件下的边坡具有较强适用性.
Slope Disaster Identification Method Based on UAV and Point Cloud Filtering
To solve the problems of high risk,high cost and vegetation coverage of traditional highway slope disaster identification method,the UAV tilt photography and machine learning technology were used to construct the point cloud filtering and disaster identification method for complex terrain environment.First,the UAV tilt photography technology was used to obtain the multi-angle high-resolution slope images,and to reconstruct the high-precision three-dimensional point cloud model of slope.Then,the support vector machine learning algorithm was used to train and establish the SVM point cloud classification model based on the morphological characteristics of highway slope surface.The slope ground and vegetation point cloud were identified and classified.The slope surface point cloud digital elevation model(DEM)was obtained.Finally,the DEM of difference(DoD)algorithm was used to detect the topographic variation of slope surface DEM obtained in different investigation periods.The slope disaster information and variation nephogram were obtained according to the detection results of 3D model change,so as to realize the automatic identification on highway slope heave,collapse,rockfall and other disasters.The technique was applied to a highway slope engineering disaster investigation in Changsha.The slope uplift area was about 621.93 m2;the average uplift height was 1.13 m;the collapse area was about 460.42 m2;the average collapse height was 0.82 m;and the total earthwork volume of slope disaster area was about 1 081.06 m3.The accuracy of proposed vegetation point cloud filtering method was over 98.4%.The result indicates that the point cloud filtering method and UAV automatic identification method based on SVM and DoD algorithm can accurately filter vegetation,quickly identify and quantify the slope disaster,and have strong applicability to the slope under complex terrain conditions.

intelligent transportslope disaster identificationUAV tilt photographysupport vector machinevariation detectionpoint cloud filtering

高文宇、余加勇、王茂枚、谢义林

展开 >

湖南大学 建筑安全与节能教育部重点实验室,湖南 长沙 410082

湖南大学 土木工程学院,湖南 长沙 410082

江苏省水利科学院研究院,江苏 南京 210017

智能交通 边坡灾害识别 无人机倾斜摄影 支持向量机 变化检测 点云滤波

湖南省水利科技项目项目湖南省自然科学基金项目湖南省科技成果转化及产业化计划项目江苏省水利科技项目

XSKJ2021000-462021JJ301022020GK20262021074

2024

10.3969/j.issn.1002-0268.2024.09.002

公路交通科技
交通运输部公路科学研究院

公路交通科技

CSTPCD北大核心
影响因子：1.007
ISSN：1002-0268
年,卷(期)：2024.41(9)
  • 9