摘要
本研究针对迭代最近点(ICP)算法在初始定位依赖性强、迭代速度慢的问题,提出一种结合内部形状描述子(ISS)特征点和改进ICP的点云配准方法。以斯坦福开源点云数据和场景点云为数据源,通过ISS算法提取关键特征点,并利用快速点特征直方图计算初始变换矩阵以实现初步配准,使得两片点云获得良好的初始位姿,最后应用基于邻域曲率优化的ICP算法完成精确配准。实验结果显示,该方法相较于传统ICP算法及基于采样一致性初始配准算法(SAC-IA)+ICP算法,在配准精度和效率上都有显著提升。
This paper addressed the strong initial position dependence and slow iterative speed of the iterative closest point(ICP)algorithm by proposing a point cloud registration method that integrated intrinsic shape signature(ISS)key points with improved ICP.The paper utilized open-source Stanford point cloud data and scene point clouds as data sources.Key features were extracted via the ISS algorithm,and initial transformation matrices were calculated by using the fast point feature histograms to achieve preliminary registration.This ensured a good initial pose for the two point clouds.Finally,the registration was refined by using an ICP algorithm based on neighborhood curvature optimization.The experimental results show that this method significantly enhances registration accuracy and efficiency compared to the traditional ICP algorithm and sample consensus initial alignment(SAC-IA)+ICP algorithm.
作者
赵永卿
ZHAO Yongqing(China Railway Engineering Design and Consulting Group Company Limited,Beijing 100055,China)
出处
《北京测绘》
2025年第2期153-157,共5页
Beijing Surveying and Mapping
基金
科技部创新工作方法专项(2020IM020500)。
关键词
点云配准
内部形态描述子
迭代最近点算法
邻域曲率
point cloud registration
intrinsic shape signature
iterative closest point algorithm
neighborhood curvature
