大型工件部分点云与整体点云的配准方法
Registration Method of Partial Point Cloud and Whole Point Cloud of Large Workpiece
-
摘要: 部分点云与整体点云的高效、高精度配准是完成大型工件尺寸快速评价工作的基础,但由于部分点云和整体点云全局特征的差异性,使用现有的局部特征描述符进行点对匹配搜索计算量大,点云配准耗时长.为此,针对部分点云与全局点云的几何特征,提出一种基于区域均值特征描述符的部分点云与整体点云配准方法.首先提出一种区域均值特征描述符,能够有效地描述点云中关键点的邻域几何特征;然后通过评价点云区域均值特征描述符的特征度选择数据点作为待配准关键点,搜索与之匹配的描述符,完成部分点云与整体点云的关键点匹配;最后使用奇异值分解法计算点云之间的转换矩阵,基于迭代最近点算法完成部分点云与整体点云的配准.利用斯坦福公共数据库点云集和大型发动机舱段的三维扫描点云数据,对配准算法的配准准确度和配准速度进行实验的结果表明,与现有的几种基于局部特征描述符(PFH,HoPPF,PPFH,FPFH)的点云配准方法相比,所提方法配准准确度平均提高56.75%,配准速度平均提高45.57%,验证了该方法的有效性.Abstract: The high-efficiency and high-precision registration of the partial point cloud and the whole point cloud is the basis for the rapid evaluation of the size of large workpieces. However, due to the difference between the global features of the partial point cloud and the whole point cloud, using the existing local feature descriptors for point pair matching search requires a lot of computation, and point cloud registration takes a long time. To solve this problem, in view of the geometric features of partial point cloud and whole point cloud, a registration method of partial point cloud and whole point cloud based on regional mean feature descriptor is proposed. Firstly, a regional mean feature descriptor is proposed, which can effectively describe the neighborhood geometric features of key points in the point cloud. Secondly, the data points are selected as the key points to be registered by evaluating the feature degree of the regional mean feature descriptors, search the matching descriptor to complete the key point matching between the partial point cloud and the whole point cloud. Finally, use the singular value decomposition method to calculate the transformation matrix between the point clouds, and register the partial point cloud and the whole point cloud based on the iterative closest point algorithm. The registration accuracy and registration speed are tested by using the point cloud set of the Stanford public database and the 3D scanning point cloud data of a large engine compartment. Compared with the point cloud registration methods of PFH, HoPPF, PPFH, and FPFH, the registration accuracy of the proposed method is increased by 56.75% on average, and the registration speed is increased by 45.57% on average. The effectiveness of the method is verified.