地球信息科学学报 ›› 2018, Vol. 20 ›› Issue (4): 452-461.doi: 10.12082/dqxxkx.2018.170634

• 全国激光雷达大会特约稿件 • 上一篇    下一篇

基于多光谱LiDAR数据的道路中心线提取

袁鹏飞1(), 黄荣刚2, 胡平波1, 杨必胜1,*()   

  1. 1. 武汉大学 测绘遥感信息工程国家重点实验室,武汉 430079
    2. 中国科学院测量与地球物理研究所 大地测量与地球动力学国家重点实验室,武汉 430077
  • 收稿日期:2017-12-25 修回日期:2018-03-25 出版日期:2018-04-20 发布日期:2018-04-20
  • 通讯作者: 杨必胜 E-mail:pfyuan1991@foxmail.com;bshyang@whu.edu.cn
  • 作者简介:

    作者简介:袁鹏飞(1991-),男,硕士生,摄影测量与遥感专业,主要从事激光雷达点云数据处理研究。E-mail: pfyuan1991@foxmail.com

  • 基金资助:
    国家自然科学基金重点项目(41531177)

Road Extraction Method Based on Multi-spectral LiDAR Data

YUAN Pengfei1(), HUANG Ronggang2, HU Pingbo1, YANG Bisheng1,*()   

  1. 1. State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, Wuhan 430079, China
    2. Institute of Geodesy and Geophysics, Chinese Academy of sciences, Geodetic and geodynamic national key laboratory, Wuhan 430077, China
  • Received:2017-12-25 Revised:2018-03-25 Online:2018-04-20 Published:2018-04-20
  • Contact: YANG Bisheng E-mail:pfyuan1991@foxmail.com;bshyang@whu.edu.cn
  • Supported by:
    National Natural Science Foundation Key Project of China, No.41531177.

摘要:

针对城市三维激光点云中,道路与地面高程相差小、激光反射强度相近使得道路提取困难;广场、停车场等地物的高程、反射强度与道路极为相近,容易产生错误提取的问题。本文设计了一种描述道路条带信息的局部二进制特征(Stripe Local Binary Feature, SLBF),结合LiDAR数据中的三维信息和多光谱信息获得基于强度、密度和平坦度等统计特征(Statistics-Based Feature, SBF),并采用随机森林分类器实现了机载点云中道路面点云和非道路面点云的有效提取。通过欧式聚类精化道路点云和迭代腐蚀边界细化中心线,进而获得矢量化的道路中心线。以Waddenzee区域的多光谱机载点云数据进行实验验证,道路中心线提取结果的完整度达到94.15%,准确度达到97.95%,精度达到92.28%。实验结果表明,该方法可以有效地提取道路中心线,同时由于设计的特征具有不变性,能够适用于城市和林间小路等各种环境。

关键词: 多光谱机载激光点云, 道路提取, 局部二进制特征, 随机森林分类器, 矢量化

Abstract:

Because of the little elevation difference between road points and ground points, and the similar laser reflection intensity between them, it is relatively hard to extract the road from lidar data at present. Furthermore, the same elevation and reflection intensity among the road, square and park makes the square and park being mistaken as road unavoidable in the city environment. In order to use the three-dimensional and multi-spectral information of the LiDAR comprehensively in this paper, data preprocessing which containing the point cloud filtering, sample collection and the data fusion is conducted first. The purpose of the filtering is to get the ground points from the LiDAR data, and the data fusion achieves the consistency of the multi-spectral LiDAR data. Then, the statistical features of the ground points can be obtained based on the intensity, the density and the flatness. To describe the road′s strip feature for distinguishing road from the square and park, the strip local binary feature (SLBF) is proposed. The SLBF is gained in a circular region which are intensity comparisons between the central position and every circular region position, and it is represented by a 96-dimension feature with value of 0 or 1. The LiDAR data is then classified as the road and non-road points by the features (Statistics-Based Feature, SBF and Stripe Local Binary Feature, SLBF) proposed above through a random forest classifier. After a further refinement by an Euclidean clustering, the road axis points are extracted by the thinning of the road points step by step by the iterative corrosion boundary method. In this paper we project the LiDAR data to the horizontal plane and use the K3M method to extract the center line of the road, and then re-project it back to the three-dimension space. Finally, the extracted road axis points are vectorized as the final result of the method. We used the multi-spectral point cloud data of the Waddenzee region to verify the method proposed in the paper. The result of the experiment shows that the completeness of the road axis vectorization achieves 94.15%, the accuracy achieves 97.95%, and the precision reaches 92.28%. The experiment shows that the proposed method can extract the road points efficiently, and vectorize the road axis correctly, it can be applied to many kinds of environments such as urban and forest as the designed features have the invariance of environments.

Key words: multi-spectral LiDAR, road extraction, local binary feature, random forest classifier, vectorization