电子测量与仪器学报

龚国强,田演,夏鑫宇.基于位姿参数估计的多视角点云配准方法[J].电子测量与仪器学报,2024,38(6):241-252

基于位姿参数估计的多视角点云配准方法

Multi-view point cloud registration method based on pose parameter estimation

DOI：

英文关键词:laser radar registration total least squares iterative closest point search algorithm

基金项目:

作者	单位
龚国强	1.三峡大学计算机与信息学院宜昌443002；2.三峡大学水电工程智能视觉检测湖北省重点实验室宜昌443002
田演	1.三峡大学计算机与信息学院宜昌443002；2.三峡大学水电工程智能视觉检测湖北省重点实验室宜昌443002
夏鑫宇	1.三峡大学计算机与信息学院宜昌443002；2.三峡大学水电工程智能视觉检测湖北省重点实验室宜昌443002

Author	Institution
Gong Guoqiang	1.College of Computer and Information Technology, China Three Gorges University, Yichang 443002, China; 2.Hubei Key Laboratory of Intelligent Vision Based Monitoring for Hydroelectric Engineering, China Three Gorges University, Yichang 443002, China
Tian Yan	1.College of Computer and Information Technology, China Three Gorges University, Yichang 443002, China; 2.Hubei Key Laboratory of Intelligent Vision Based Monitoring for Hydroelectric Engineering, China Three Gorges University, Yichang 443002, China
Xia Xinyu	1.College of Computer and Information Technology, China Three Gorges University, Yichang 443002, China; 2.Hubei Key Laboratory of Intelligent Vision Based Monitoring for Hydroelectric Engineering, China Three Gorges University, Yichang 443002, China

摘要点击次数: 37

全文下载次数: 1273

中文摘要:

传统的点云配准算法通过两点云数据之间的特征实现对应点配对，这种方法要求点云具有明确的特征，且存在计算量大、匹配时间长、配准精度低等问题，而ICP算法虽然应用广泛，但对初始值敏感。对此，提出了一种基于位姿参数估计的多视角点云配准方法(PPE-ICP)。首先通过分析误差的分布特性可证明误差极小值存在，使用A*搜索算法寻找误差极小值，降低误差传播的影响，为后续的参数估计提供较好的初值；其次将总体最小二乘估计引入点云配准，在不依赖点云数据的同时，使用少量参考点就能获得点云从目标坐标系到东北天坐标系的转换矩阵，完成点云位姿矫正，结合迭代最近点算法(ICP)，实现点云精确配准。通过与FGR-ICP、FPFH-ICP、NDT-ICP、RANSAC-TrICP和KSS-ICP这5种方法在公开数据集和自制实验装置收集到的点云上进行对比实验，点云数据量为20 000点时实现配准只需6.55 s，极大地降低了大数据量下点云配准的时间成本，在实地点云配准中平移误差最大不超过0.03 m，旋转误差控制在0.07°。实验结果表明，PPE-ICP对相似变换、残缺点云和低重复率具有较强的鲁棒性，在多视角点云配准中具有较高的配准效率和配准精度。

英文摘要:

The traditional point cloud registration algorithm achieves corresponding point pairing through features between two-point cloud datasets. This method requires point clouds to possess distinct features, yet it suffers from issues such as high computational complexity, long matching time, and low registration accuracy. Although the ICP algorithm is widely used, it is sensitive to initial values. To address these challenges, we propose a multi-view point cloud registration method based on pose parameter estimation (PPE-ICP). Firstly, by analyzing the distribution characteristics of errors, we demonstrate the existence of error minima. The A* search algorithm is then employed to locate these minima, reducing the impact of error propagation and providing improved initial values for subsequent parameter estimation. Secondly, we introduce total least squares estimation into point cloud registration, which, without relying on point cloud data, utilizes a limited number of reference points to obtain the transformation matrix from the target coordinate system to the Northeast-Up (ENU) coordinate system. This accomplishes point cloud pose correction, and in combination with the Iterative Closest Point (ICP) algorithm, achieves precise point cloud registration. Comparative experiments were conducted with five methods: FGR-ICP, FPFH-ICP, NDT-ICP, RANSAC-TrICP, and KSS-ICP, using both publicly available datasets and point clouds collected from a self-made experimental setup. When dealing with a point cloud dataset of 20 000 points, our PPE-ICP achieves registration in just 6.55 seconds, significantly reducing the time cost for point cloud registration with large datasets. In field applications, the maximum translation error is less than 0.03 m, and the rotation error is controlled within 0.07°. The experimental results demonstrate that PPE-ICP exhibits strong robustness against similar transformations, incomplete point clouds, and low repetition rates, achieving high registration efficiency and accuracy in multi-view point cloud registration.

查看全文查看/发表评论下载PDF阅读器