基于K-D树加速的大型点云配准算法

Large-scale Point Cloud Registration Algorithm Based on K-D Tree Acceleration

为了实现大型点云的精确配准,首先对大型点云数据进行降采样及去质心预处理,并采用迭代最近点(Iterative Closest Point,ICP)算法计算点云间的旋转矩阵R与平移矩阵T,最终在Visual Studio中基于OpenGL库实现界面交互和结果显示。结果表明,该配准算法既支持人为选择关键点进行配准,也支持随机生成关键点进行配准,且准确率高。大型点云数据实验表明,针对大型点云点数目多,普通配准算法计算时间长的问题,采用K-D树进行配准算法加速,可保证精度,同时大大缩短了大型点云配准计算时间。

To achieve the precise registration of large point cloud,down-sampling and de-centroid preprocessing of large point cloud data are conducted first,and then the iterative closest point(ICP) algorithm is used to calculate the rotation matrix R and translation matrix T between the point clouds. The interface interaction and result display are finally achieved based on OpenGL library in Visual Studio. The results show that the registration algorithm not only supports manual selection of key points for registration,but also supports random generation of key points, and the registration accuracy is high. It is shown in the experiments that compared with the traditional registration algorithm, K-D tree accelerated registration algorithm not only guarantees the accuracy but also greatly reduces the computational time of registration for large-scale point cloud with many points.