汽车工程 ›› 2023, Vol. 45 ›› Issue (4): 572-578.doi: 10.19562/j.chinasae.qcgc.2023.04.005

所属专题: 智能网联汽车技术专题-感知&HMI&测评2023年

• • 上一篇    下一篇

双重下采样增强的点云改进配准算法研究

陈仲生1,2(),李潮林2,左旺2,侯幸林1   

  1. 1.常州工学院汽车工程学院,常州  213032
    2.湖南工业大学电气与信息工程学院,株洲  417002
  • 收稿日期:2022-10-16 修回日期:2022-11-20 出版日期:2023-04-25 发布日期:2023-04-19
  • 通讯作者: 陈仲生 E-mail:chenzs@czu.cn
  • 基金资助:
    国家自然科学基金(62101074)

Study on Improved Point Cloud Registration Algorithm Enhanced by Double Down-sampling

Zhongsheng Chen1,2(),Chaolin Li2,Wang Zuo2,Xinglin Hou1   

  1. 1.College of Automotive Engineering,Changzhou Institute of Technology,Changzhou  213032
    2.College of Electrical and Information Engineering,Hunan University of Technology,Zhuzhou  417002
  • Received:2022-10-16 Revised:2022-11-20 Online:2023-04-25 Published:2023-04-19
  • Contact: Zhongsheng Chen E-mail:chenzs@czu.cn

摘要:

自动驾驶领域对点云配准实时性要求高,而已有ICP算法及其变体存在对初始位姿要求高、配准速度慢等问题。鉴于此,本文提出了一种改进的快速点云配准方法,首先采用双重下采样方法对初始点云数据进行预处理,在保留原始特征的同时快速降低点云数据量,然后引入内部形状描述子(ISS)来优化超级全等四点集(Super4PCS)算法,降低其时间复杂度,最后选用线性最小二乘优化ICP算法进行快速精配准。采用斯坦福点云数据和自动驾驶Kitti点云数据对该算法有效性进行了测试和对比验证,结果表明:该算法具有良好的鲁棒性,且配准精度和配准速度均比已有算法有明显提高。

关键词: 点云配准, 双重下采样, ISS特征, Super4PCS, ICP算法

Abstract:

In the field of autonomous driving, the real-time requirement of point cloud registration is high. Existing ICP algorithm and its variants have such problems as high requirements for initial pose and slow registration speed. In order to deal with the above-mentioned problems, an improved fast point cloud registration algorithm is proposed in this paper. Firstly, a double down-sampling method is used to preprocess point cloud data. By this way, the amount of point cloud data can be greatly reduced rapidly, while retaining the original features. Then the Intrinsic Shape Signature (ISS) is introduced to optimize the Super 4-Points Congruent Sets (Super4PCS) algorithm to reduce its time complexity. Finally, the linear least squares optimization-based iterative closest point (ICP) algorithm is used for fast and precise registration. The effectiveness of the algorithm is tested and compared by using the Stanford and autonomous driving Kitti point cloud data. The results show that the proposed algorithm has good robustness, and the registration accuracy and speed are significantly increased compared with the existing algorithms.

Key words: point cloud registration, double down-sampling, ISS, Super4PCS, ICP