平面参数空间的实时三维点云配准方法  

A method for real-time 3D point cloud registration in a parameter space of planes

作　　者：王力宇[1] 曹其新[1] 王雯珊[1] Wang Liyu Cao Qixin Wang Wenshan(State Key Laboratory of Mechanical System and Vibration, Research Institute of Robotics, Shanghai Jiaotong University, Shanghai 20024)

机构地区：[1]上海交通大学机器人研究所机械系统与振动国家重点实验室,上海200240

出　　处：《高技术通讯》2017年第4期351-358,共8页Chinese High Technology Letters

基　　金：国家自然科学基金(61273331)资助项目

摘　　要：针对具有多平面结构的室内环境的三维定位和环境建模,提出了一种在平面参数空间进行配准的实时三维点云配准方法。该方法首先使用一种改进的三维霍夫变换算法快速提取点云中的平面特征,然后使用迭代算法在平面的参数空间中寻找最近平面,最后使用这些平面的对应关系来估计两帧点云之间的位姿变换关系。在一个实验室场景中进行的对比实验表明,该算法能够达到与传统迭代最近点(ICP)算法相似的精度,而且速度大大提升。在仅使用普通笔记本CPU的情况下即可实现实时的点云拼接。For 3 D localization and modelling of an indoor environment with a multi-plane structure, a new method for real-time 3D point cloud registration in a planar parameter space was proposed. The method uses an improved 3D hough transform to quickly extract the plane features in point clouds, then, searches the closest planes in the parameter space of the planes by using an iterative method, and finally, estimates the translation and rotation of the point clouds by registration of the correspoding planes. A comparison experiment performed in a laboratory environ- ment showed that the registration accuracy of the proposed method reached the traditional iterative closest point （ICP） method, and the registration speed was greatly improved. The new method can achieve the real-time registration of point clouds when using an ordinary laptop CPU.

关 键 词：点云配准 平面提取 霍夫变换 三维地图重建 迭代最近点(ICP) 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]