平面特征约束下的多视点云配准方法  

Multi-view point cloud registration method based on planar feature

作　　者：党空雁 魏冠军 吴志才 赵丰炯 DANG Kongyan;WEI Guanjun;WU Zhicai;ZHAO Fengjiong(Faculty of Geomatics,Lanzhou Jiaotong University,Lanzhou 730070,China;National-Local Joint Engineering Research Center of Technologies and Applications for National Geographic State Monitoring,Lanzhou 730070,China;Gansu Provincial Engineering Laboratory for National Geographic State Monitoring,Lanzhou 730070,China)

机构地区：[1]兰州交通大学测绘与地理信息学院,兰州730070 [2]地理国情监测技术应用国家地方联合工程研究中心,兰州730070 [3]甘肃省地理国情监测工程实验室,兰州730070

出　　处：《测绘科学》2023年第4期192-198,230,共8页Science of Surveying and Mapping

基　　金：国家自然科学基金项目(41964008)。

摘　　要：针对由于多视点云的密度不同,将同名特征点作为配准基元的点云配准方法无法找到具有精准对应关系的点对,从而存在配准精度不高的问题,该文提出一种基于同名特征平面的点云配准方法。该方法将坐标原点在同名特征平面上的投影点作为同名特征点,以空间点面关系具有旋转不变性为约束条件,引入对偶四元数描述空间变换参数,基于最小二乘准则构建目标函数,利用Levenberg-Marquardt法解决配准模型的非线性优化问题。最后通过实测数据实验验证算法的正确性与有效性。结果表明:该方法能够实现实际场景中建/构筑物的多视点云配准;采用Levenberg-Marquardt法在迭代过程中可有效避免局部最小陷阱;对偶四元数有效减少了解算空间变换参数中的耦合误差。Due to the different density of point clouds in different perspectives,the point cloud registration method that uses the feature points of the same name as the registration primitives cannot find point pairs with accurate correspondence,so the registration accuracy is not high.In this paper,a point cloud registration method based on the feature plane of the same name was proposed.The projection point of the coordinate origin on the feature plane of the same name were taken as the feature point of the same name,and the spatial point-plane relationship with rotational invariance was taken as a constraint;then dual quaternions was used to describe spatial transformation parameters,the objective function was constructed based on the least squares criterion,and the nonlinear optimization problem of the registration model was solved by the Levenberg-Marquardt method.Finally,the correctness and effectiveness of the algorithm were verified by the experimental data.The results showed that the proposed method could realize point cloud registration under different viewing angles of buildings in actual scenes;the Levenberg-Marquardt method could effectively avoid the local minimum trap in the iterative process,and the dual quaternion could effectively reduce the transformation parameters of the solution space coupling error.

关 键 词：点云配准 平面特征 对偶四元数 空间相似变换 

分 类 号：P225.2[天文地球—大地测量学与测量工程]