基于点到面度量的多视角点云配准方法  

Registration of Multi-View Point Sets Based on Point-to-Plane Measurement

作　　者：马洁莹 田暄 翟庆 王丞 MA Jieying;TIAN Xuan;ZHAI Qing;WANG Cheng(School of Software Engineering,Xi’an Jiaotong University,Xi’an 710049,China)

机构地区：[1]西安交通大学软件学院,西安710049

出　　处：《西安交通大学学报》2022年第6期120-132,共13页Journal of Xi'an Jiaotong University

基　　金：陕西省重点研发计划项目(2021GY-025)。

摘　　要：针对现有的大多数基于点到点度量的多视角点云配准方法在配准过程中,由于物体表面离散化而无法获得点到点的精确重叠,从而导致的收敛速度慢、配准精度低的问题,提出一种基于点到面度量的多视角点云配准方法。为获得多视角点云匹配结果,采用逐步求精的策略将多视角配准问题分解成多个点到面双视角配准子问题进行求解。在双视角配准过程中:使用数据点与对应点处切平面的距离误差代替点对距离误差,给出新的目标函数;提出高效法向量转换策略,以减少多视角配准的每次迭代中平面法向量的求解次数。在目标函数的求解过程中,用线性最小二乘法逼近非线性优化问题,从而实现点到平面误差的最小化。将所提方法在斯坦福数据集上进行了测试,实验结果表明:与当下较为流行的多视角配准方法相比,所提方法在不同数据集上的旋转误差均降低了38.9%以上,平移误差均降低了16.6%以上,能够快速实现精确、可靠的多视角点云配准。Most of the existing multi-view point set registration methods based on point-to-point measurement cannot obtain the accurate overlap of the individual points due to the discretization of object surface,resulting in the problems of slow convergence and low registration accuracy.To solve this problem,this paper proposed a multi-view point set registration method based on point-to-plane measurement.To achieve the multi-view registration,the proposed method uses the stepwise refinement strategy to break down the multi-view registration problem into several point-to-plane pair-wise registration problems.In the pair-wise registration process,first of all,a new objective function is given which uses the point-to-plane error to replace the point-to-point distance error;Secondly,an effective normal vector transformation strategy is proposed to reduce the times of solving the plane normal vector in each iteration of the multi-view point set registration.Then,in the process of solving the objective function,the linear least square method is used to approximate the nonlinear optimization problem,so as to minimize the point-to-plane error.Finally,the proposed method is tested on Stanford data sets,and the experimental results show that:compared with four other multi-view registration methods,the proposed method reduces the rotation error by more than 38.9%in terms of different data sets,and the translation error by more than 16.6%,thus effectively achieving accurate and reliable multi-view point sets registration.

关 键 词：多视角配准 点到面度量 线性最小二乘 平面法向量 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]