地球信息科学学报 ›› 2018, Vol. 20 ›› Issue (4): 471-479.doi: 10.12082/dqxxkx.2018.170423

• 全国激光雷达大会特约稿件 • 上一篇    下一篇

回波强度约束下的无人机LiDAR点云K-means聚类滤波

李沛婷(), 赵庆展*(), 陈洪   

  1. 1. 石河子大学信息科学与技术学院,石河子 832000
    2. 国家遥感中心新疆兵团分部,石河子 832000
    3. 兵团空间信息工程技术研究中心,石河子 832000
  • 收稿日期:2017-09-12 修回日期:2018-01-16 出版日期:2018-04-20 发布日期:2018-04-20
  • 通讯作者: 赵庆展 E-mail:sw_lpt@sina.com;zqz_inf@shzu.edu.cn
  • 作者简介:

    作者简介:李沛婷(1993-),女,硕士生,主要从事空间信息技术及应用、无人机遥感应用基础、点云数据处理研究。E-mail: sw_lpt@sina.com

  • 基金资助:
    新疆生产建设兵团科技计划项目(2015BA006);兵团空间信息工程技术研究中心创建项目(2016BA001);中央引导地方科技发展专项资金项目(201610011)

Filtering UAV LiDAR Point Cloud with K-means Clustering under the Constraint of Echo Intensity

LI Peiting(), ZHAO Qingzhan*(), CHEN Hong   

  1. 1. College of Information Science and Technology, ShiHeZi University, Shihezi 832000, China
    2. Division of National Remote Sensing Center, Xinjiang Production and Construction Corps, Shihezi 832000, China
    3. Geospatial Information Engineering Research Center, Xinjiang Production and Construction Corps, Shihezi 832000, China
  • Received:2017-09-12 Revised:2018-01-16 Online:2018-04-20 Published:2018-04-20
  • Contact: ZHAO Qingzhan E-mail:sw_lpt@sina.com;zqz_inf@shzu.edu.cn
  • Supported by:
    Xinjiang Production and Construction Corps Science and Technology Project, No.2015BA006;Geospatial Information Engineering Research Center To Create, Xinjiang Production and Construction Corps, No.2016BA001;The Central Government Directs Local Science and Technology Development Special Funds, No.201610011.

摘要:

针对目前无人机激光雷达点云滤波过程中存在的效率低、误分割和精度差等问题,本文在对点云三维坐标进行K-means聚类得到不同聚类结果的基础上,引入最大-最小标准化方法对不同聚类结果的点云回波强度原始值进行标准化,得到 0-1范围回波强度规则值。针对不同聚类结果选择不同范围的回波强度规则值得到对应地面点云,以提升研究区点云的滤波精度并减少其地面点云的数据量。同时,对比利用K-means聚类对三维坐标和回波强度原始值进行滤波的结果。结果表明:对研究区点云去噪、抽稀预处理后得到107 372个点云数,利用K-means方法对三维坐标和回波强度原始值进行聚类滤波得到地面点数为66 713个,占点云总数的62.133%。通过使用本文方法可剔除过分割地表植被点13 648个,得到地面点云占点云总数的49.422%。该方法能够较好地保持地形轮廓并降低地面点云的数据量,从而为后期快速建立高精度DEM奠定基础。

关键词: 无人机LiDAR, K-means聚类, 点云预处理, 点云滤波, 回波强度标准化

Abstract:

Digital elevation model (DEM) is an effective way of describing terrain. Unmanned Aerial Vehicle (UAV) light detection and ranging (LiDAR) has become a novel and powerful technology to produce DEM. Point cloud filtering is very important to generate DEM. However, low efficiency, over-segmentation, under-segmentation, and low precision have been problems in point cloud filtering. In order to improve filter accuracy and reduce ground point clouds data volume for rapidly establishing high accuracy DEM, this paper puts forward a filter method by K-means clustering to acquire ground point clouds based on constraint of normalized echo intensity values. UAV Scout B1-100 was used to carry laser scanner VUX-1 to acquire point clouds with high density and resolution in Xinjiang mana’s valley. The Riegl LMS and OxTS NAVgraph software were then used to carry out registration and correction of point clouds in study area. The point clouds after processing of removing noise points and diluting points had a total of 107 372 points. Then, K-means method was used under constraint of three-dimensional coordinates’ values of point clouds to get three different clustering results. Meanwhile, maximum-minimum standard method was introduced to normalize values of original echo intensity to a range of 0 to 1. The corresponding ground point clouds were obtained for different clustering results by choosing different ranges based on normalized intensity values. Finally, we merged ground point clouds from different clustering results to acquire ground point clouds in entire study area. For comparisons, we also used K-means clustering method under constraint of original values of echo intensity and three-dimensional coordinates. The results show that the ground point cloud obtained from K-means clustering and constraint with original values ??of echo intensity has 66 713 points, accounting for 62.133% of the total number of point clouds in this paper. An additional 13 648 subsurface vegetation points can be removed by using this paper’ method, reducing the ratio of ground point clouds to total point clouds to 49.422%. This method can better maintain terrain profile and reduce data volume of the ground point cloud, thus laying the foundation for the rapid establishment of high precision DEM.

Key words: UAV LiDAR, K-means clustering, point cloud preprocessing, point cloud filtering, echo intensity normalization