一种面向扭锁识别的多视角三维点云配准方法
DOI:
CSTR:
作者:
作者单位:

1.上海海事大学物流工程学院上海201306;2.上海海事大学高级技术学院上海201306; 3.上海海瞩智能科技有限公司上海201306;4.里斯本大学学院技术与建筑学院里斯本1649-026

作者简介:

通讯作者:

中图分类号:

TP391.4;TN958.98

基金项目:

国家自然科学基金(52472435)、上海科学技术委员会(25692109300)、上海市教育科学项目(B2023003)、上海市教育委员会(人工智能促进科研范式改革赋能学科跃升计划)、国家留学基金管理委员会(202507920006)项目资助


Multi-view 3D point cloud registration method for identifying twist locks
Author:
Affiliation:

1.Logistics Engineering College, Shanghai Maritime University, Shanghai 201306, China; 2.Higher Technical College, Shanghai Maritime University, Shanghai 201306, China; 3.Shanghai SMUVision Smart Technology Ltd., Shanghai 201306, China; 4.School of Technology and Architecture, ISCTE-Instituto Universit-rio de Lisboa, Lisbon 1649-026, Portugal

Fund Project:

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
  • |
  • 文章评论
    摘要:

    针对港口自动化装卸作业中集装箱扭锁锁柄精准识别与位姿估计的需求,提出一种基于多视角的点云局部配准方法。首先对采集的点云进行统计滤波与体素下采样预处理。为解决多视角点云间非重叠区域造成的结构性干扰,采用基于密度的噪声应用空间聚类(density-based spatial clustering of applications with noise, DBSCAN)聚类算法提取两片点云的最大重叠区域即锁孔区域作为配准基准,并通过参数分析选取最优邻域半径和最小点数。在此基础上采用由粗到精的配准策略,基于快速点特征直方图(fast point feature histograms, FPFH)特征的随机采样一致性(random sample consensus, RANSAC)算法实现粗配准,再结合引入自适应距离阈值与法线一致性约束的改进迭代最近点(iterative closest point,ICP)算法进行精配准,其中自适应阈值根据锁孔几何先验动态演化,法线约束有效剔除边缘误匹配。实验结果表明,以锁孔区域为基准的局部配准相较于全局配准,配准精度提升约62%,时间效率提高约59%;在噪声环境下改进ICP算法表现出良好鲁棒性。最终将配准后点云应用于锁柄识别,三维重建尺寸误差显著降低,验证了该方法为锁柄精准识别提供有效数据支持,在提升港口自动化作业精度与效率方面具有实际意义。

    Abstract:

    This paper proposes a multi-view point cloud local registration method to address the need for accurate identification and pose estimation of container twist lock handles in automated port loading and unloading operations. First, statistical filtering and voxel downsampling are performed on the acquired point clouds. To address structural interference caused by non-overlapping regions between multi-view point clouds, the density-based spatial clustering of applications with noise (DBSCAN) clustering algorithm is used to extract the region of maximum overlap between the two point clouds—specifically, the lock hole region—as the registration reference. Optimal neighborhood radii and minimum point counts are selected through parameter analysis. Based on this, a coarse-to-fine registration strategy is adopted: the random sample consensus (RANSAC) algorithm using fast point feature histograms (FPFH) features is employed for coarse registration, followed by fine registration using an improved iterative closest point (ICP) algorithm that incorporates adaptive distance thresholds and normal consistency constraints. The adaptive thresholds dynamically evolve based on the geometric prior of the lock hole, while the normal constraints effectively eliminate edge misalignments. Experimental results show that local registration using the lock hole region as a reference improves registration accuracy by approximately 62% and increases time efficiency by approximately 59% compared to global registration; the improved ICP demonstrates good robustness in noisy environments. Finally, the registered point cloud was applied to lock handle recognition, resulting in a significant reduction in 3D reconstruction dimensional errors. This validates that the method provides effective data support for accurate lock handle recognition and has practical significance for enhancing the accuracy and efficiency of port automation operations.

    参考文献
    相似文献
    引证文献
引用本文

王腾,沈阳,张钰洁,崔铭杰,张志伟,宓超.一种面向扭锁识别的多视角三维点云配准方法[J].电子测量与仪器学报,2026,40(9):89-100

复制
分享
相关视频

文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:
  • 最后修改日期:
  • 录用日期:
  • 在线发布日期: 2026-11-13
  • 出版日期:
文章二维码
×
《电子测量与仪器学报》
关于防范虚假编辑部邮件的郑重公告