一种基于退火机理的高精度自动点云配准算法

A High-Precision Automatic Point Cloud Registration Algorithm Based on Annealing Mechanism

高精度器件装配是制造业的重要需求,采用数字化手段辅助配准可以更清楚发现模型的配合关系。针对点云配准处理过程中配准精度低且耗时长等问题,本文提出一种基于退火机理结合特征面与特征参数的高精度自动点云配准算法。首先根据曲率值的大小排序并选最小的点作为种子点,从最平坦的区域依据区域生长准则进行特征提取,针对分割出的特征面使用快速点特征直方图(Fast Point Feature Histograms,FPFH)描述并寻找点对关系。随机采样一致性算法排除错误点对,奇异值分解法求取变化矩阵完成粗配准,最后引入退火机理的思想改变特征参数,通过迭代找到全局最优解完成点云配准。通过对3D Match点云数据中的圆柱体进行点云配准,与传统的迭代最近点(Iterative Closest Point,ICP)云配准方法、四点全等集合算法+ICP、随机采样一致性初始配准算法+ICP方法比较,本算法提供了较为理想的初始位置,有效提高了配准的精度与速度,可对高精度配准提供重要支撑。

High-precision device assembly is an important requirement in the manufacturing industry,and the use of digital means to assist in registration can help to clearly identify the relationship between models.In order to solve the problems of low registration accuracy and long time in the process of point cloud registration,a high-precision automatic point cloud registration algorithm combining feature surfaces and feature parameters.Firstly,sort according to the size of the curvature value and choose the smallest one.The point is used as the seed point,and the feature extraction is carried out from the flattest area according to the regional growth criterion,and it is quickly used to segment the feature surface.Fast Point Feature Histograms(FPFH)are used to describe and look for point-pair relationships.The Random Sample Consensus algorithm is used to eliminate incorrect point pairs,and the Singular Value Decomposition method is applied to obtain the transformation matrix for rough registration.Finally,by introducing the annealing mechanism to adjust feature parameters,the algorithm iterates to find the global optimal solution for point cloud registration.By registering a cylinder in 3D Match point cloud data,this algorithm is compared with the traditional Iterative Closest Point(ICP)registration method,Four Points Congruent Sets+ICP,and Sample Consensus Initial Alignment+ICP methods.The algorithm provides an ideal initial position,effectively improving the accuracy and speed of registration,and can provide important support for high-precision registration.