三维点云的邻域分布快速配准

打开文本图片集

关键词：点云配准；邻域分布特征；匹配点对优化；迭代最近点；体素化网格法中图分类号：TP391.41 文献标识码：Adoi：10.37188/OPE.20253319.3106 CSTR：32169.14.OPE.20253319.3106

Abstract：A fast point-cloud registration method based on neighborhood distribution features is proposed to address the high computational cost of traditional high-dimensional feature extraction and the slow performance of dense registration algorithms that rely on coarse-fine two-step feature matching. First，three deep geometric features of neighboring points are defined，and a low-dimensional，multi-scale neighbor hood distribution descriptor is introduced to substantially reduce feature-computation complexity while enhancing descriptor discriminability for efficient characterization of local point-cloud properties. Second，a rapid coarse-registration scheme is developed using the neighborhood distribution descriptor： feature points are selected according to the global undulation degree and neighborhood distribution direction；initial correspondences are established based on the neighborhood distribution descriptor；and Euclidean-distance constraints between point pairs are strengthened to remove incorrect matches，enabling eficient and accurate coarse alignment. Finally，to accelerate dense registration，the iterative closest point （ICP） algorithm is improved using a Δk -dimensional tree and voxel-grid downsampling，and a quadratic fine-registration strategy is employed to correct downsampling-induced errors，thereby further improving fine-registration accuracy and eficiency. Experiments on Stanford models and industrial part point clouds demonstrate that，compared with conventional feature-descriptor-based methods，the proposed approach increases registration ac- curacy by over 22% and reduces computation time by more than 43% ，confirming its effectiveness，robustness，and practical applicability for rapid registration of object-surface point clouds acquired from different viewpoints.

Key words： point cloud registration； neighborhood distribution feature；matched point pair optimization; iterative closest point;voxelized grid method

1引言

点云配准作为三维点云处理技术中的重要部分，广泛应用于工业检测[1]、三维测绘[2]、文物数字化3等领域。（剩余16345字）