摘要
针对多平面结构的物体,传统的点特征点云配准方法存在鲁棒性差、易收敛到局部最优解等问题,提出了一种基于法向量投票的点云配准方法。用平面特征代替点特征作为配准基元,建立基于平面的坐标转换模型。首先构建kd-tree,计算各点的法向量,并将法向量转换到霍夫空间进行投票,提取平面特征;然后将单位四元数作为特征描述算子,以同名平面特征作为约束条件,根据最小二乘平差原则,求解点云之间的位姿变换关系。实验结果表明:相较于其他两种方法,提出方法对初始位置没有依赖性,在配准过程中可以有效避免局部最小陷阱,并且配准精度得到了提高。
Aiming at the problems of poor robustness and easy convergence to local optimal solution of traditional point cloud registration methods for objects with multi-planar structures,this paper proposed a point cloud registration method based on normal vector voting.It used planar features instead of point features as registration primitives to establish a coordinate transformation model.Firstly,it constructed kd-tree,and calculated the normal vector of each point,converted the normal vector to the Hough space and voted to extract the planar features.Then,it used the unit quaternions as feature description operators,and used conjugate planes as constraints.According to the least square adjustment principle,it solved the pose transformation relationship between point clouds.Experimental results show that,compared with the other two methods,the proposed method does not depends on the initial position,and can effectively avoid local minimum trapping in the registration process,and improves the registration accuracy.
作者
周颖
林意
Zhou Ying;Lin Yi(School of Artificial Intelligence&Computer Science,Jiangnan University,Wuxi Jiangsu 214122,China)
出处
《计算机应用研究》
CSCD
北大核心
2022年第2期637-640,共4页
Application Research of Computers
关键词
点云配准
法向量
平面特征
四元数
point cloud registration
normal vector
plane feature
quaternion
