计量学报2025，Vol.46Issue(6)：830-838,9.DOI:10.3969/j.issn.1000-1158.2025.06.07
基于SIFT-3DSC的GICP三维点云配准技术研究
Research on 3D Point Cloud Registration Technology of GICP Based on SIFT-3DSC
摘要
Abstract
A registration algorithm combining the generalized iterative closest point(GICP)optimization with the scale-invariant feature transform descriptors and 3D shape context features(3D-SIFT)is proposed.Firstly,actual adhesive workpieces from automotive adhesive application production lines are selected as the experimental subjects.A combined type point cloud data acquisition system is constructed to scan these workpieces,different types of colloid point cloud data are obtained.After that,preprocessing steps,including plane segmentation,noise filtering,and voxel downsampling,are applied to the acquired data to obtain moderate-sized point clouds containing only adhesive elements.Initial rough registration is performed using the Kalman filter point cloud registration with 3D-SIFT algorithm,generating the initial transformation matrix.Finally,the GICP algorithm is utilized to refine the registration for three types of adhesive points—disconnected,fine,and overflowed—against standard point cloud models,yielding precise pose transformation matrices for each point cloud group.To validate the effectiveness of the proposed method,the registration results are compared with those of alternative algorithms,including SAC-IA+ICP and SAC-IA+NDT.Root mean square error and computational time are evaluated.The experimental results demonstrate a 15.3%and 18.25%improvement in registration accuracy over the SAC-IA+ICP and SAC-IA+NDT algorithms,respectively,with time savings of 13.31%and 49.91%.关键词
视觉检测/GICP优化算法/3D-SIFT/点云配准技术/卡尔曼滤波配准算法Key words
visual inspection/GICP optimization algorithm/3D-SIFT/point cloud registration technology/Kalman filter registration algorithm分类
通用工业技术引用本文复制引用
李卓妍,胡晓峰,范伟军,郭斌,潘飞文,罗哉..基于SIFT-3DSC的GICP三维点云配准技术研究[J].计量学报,2025,46(6):830-838,9.基金项目
国家重点研发计划(2022YFF0705704,2022YFF0705705) （2022YFF0705704,2022YFF0705705）
浙江省"尖兵"计划项目(2023C01061) （2023C01061）
