自研激光雷达三维点云配准技术

为了实现激光雷达点云与图像重建点云的三维空间配准，基于自研三维扫描激光雷达系统，提出了新型的快速多尺度因子（FMSR）点云配准算法，研究了空间点云配准技术。该算法主要包括初始配准和精确配准2个步骤:初始配准使用基于尺度自适应关键点质量（ASKQ）的点云特征提取算法，提取关键点的特征匹配对，求解点云配准初始参数；精确配准利用K-邻近（KNN）算法全局搜索，提升计算效率，多次迭代得到2组点云之间的最优旋转矩阵、最优平移向量和最优尺度因子。仿真和实验结果表明，所提出的算法对空间目标（尺寸为20.30 m×7.85 m×26.56 m）实现空间点云配准，配准精度达到0.194 m，运行时间为16.207 s；与多尺度迭代最近点（S-ICP）算法相比，配准精度提高了0.131 m，运行时间提高了30%。所提出的空间点云配准技术可为场景重建和纹理匹配提供算法基础。 

Three-dimensional point cloud registration technique for self-designed LiDAR

In order to realize the registration of point cloud data respectively obtained from LiDAR and camera, we used a fast multi-scale registration (FMSR) algorithm to register the point cloud data, based on a self-designed three-dimensional scanning laser radar system in our laboratory. The algorithm includes two steps:coarse registration and fine registration. In the coarse registration, an adaptive scale key point quality (ASKQ) algorithm was used to match key points and determine the initial parameters for fine registration. And in the fine registration, K-nearest neighbors (KNN) algorithm was used to simplify the search process and improve the algorithm efficiency. The optimal rotation matrix, translation vector and scale factor between two sets of point cloud data were obtained through many iterations. The simulation verified the stability of FMSR algorithm for multiscale registration. Simulation and experimental results show that the proposed algorithm successfully registers the point cloud data of the self-made LiDAR system and commercial camera. The root-mean-square error of the registration is 0.194 m and the execution time is 16.207 s, for a building with size of 20.30 m×7.85 m×26.56 m. Compared with an existing scale-iterative closest point (S-ICP) algorithm, the registration accuracy of the proposed algorithm is improved by 0.131 m, and the execution time is reduced by 30%. The proposed point cloud registration method can provide an algorithm basis for scene reconstruction and texture matching.