结合特征点提取的点云配准算法

Point Cloud Registration Algorithm Combined with Feature Points Extraction

摘要: 在点云配准技术中，粗、精配准的策略被用作点云配准的常用手段，本文针对普通配准策略配准耗时长，配准精度有待提高的问题，提出了一种耗时更短、精度更高的点云配准算法。首先，使用SIFT算法提取源、目标点云的特征点，将这些特征点作为配准算法的输入；然后，使用SAC-IA算法进行点云粗配准，为后续精配准提供一个大致对齐的位姿；最后，使用带法向量约束的点到面ICP算法进行点云精配准，得到最终的配准位姿。实验表明，本文所提算法在配准耗时上相比于SAC-IA + ICP算法提升了96.2%、相比于NDT + ICP算法提高了81.0%，在配准的均方根误差上相比于SAC-IA + ICP算法提高了43.6%、相比于NDT + ICP算法提高了24.7%，证明了本文算法在计算时间和配准精度上的有效性。

Abstract: In view of the problem that the registration of ordinary registration strategies takes a long time and the registration accuracy needs to be improved, this paper proposes a point cloud registration algorithm with shorter time and higher accuracy. Firstly, the SIFT algorithm is used to extract the feature points of the source and target point clouds, and these feature points are used as the input of the registration algorithm, then the SAC-IA algorithm is used for the rough registration of the point cloud to provide a roughly aligned pose for the subsequent fine registration, and finally, the point-to-surface ICP algorithm with normal vector constraint is used for the point cloud fine registration to obtain the final registration pose. Experiments show that the proposed algorithm is 96.2% longer than the SAC-IA + ICP algorithm in terms of registration time, 81.0% higher than the NDT + ICP algorithm, 43.6% higher than the SAC-IA + ICP algorithm and 24.7% higher than the NDT + ICP algorithm in the root mean of registration square error, which proves the effectiveness of the proposed algorithm in terms of calculation time and registration accuracy.