基于工业三维检测的点云配准技术研究进展  

Research progress of point cloud registration technology based on industrial 3D inspection

作　　者：王水仙 邓朝晖 葛吉民[1,3] 刘伟 WANG Shuixian;DENG Zhaohui;GE Jimin;LIU Wei(Institute of Intelligent Manufacturing,Hunan University of Science and Technology,Xiangtan 411100,Hunan,China;Manufacturing Engineering Research Institute,Huqiao University,Xiamen 361021,Fujian,China;Hunan Provincial Key Laboratory of High Efficient and Precision Machining of Difficult-to-Cut Materials,Hunan University of Science and Technology,Xiangtan 411100,Hunan,China)

机构地区：[1]湖南科技大学机电工程学院,湖南湘潭411100 [2]华侨大学制造工程研究院,福建厦门361021 [3]湖南科技大学,难加工材料高效精密加工湖南省重点实验室,湖南湘潭411100

出　　处：《金刚石与磨料磨具工程》2023年第3期285-297,共13页Diamond & Abrasives Engineering

基　　金：湖南省自然科学省市联合基金(2021JJ50116);湖南省高新技术产业科技创新引领计划(2020GK2003);国家自然科学基金−浙江两化融合联合基金(U1809221)。

摘　　要：随着制造业的发展,所需零件逐渐向大尺寸、复杂形状、表面加工质量高等方向发展,且在加工过程中对零件质量进行检测是必不可少的环节。为提高质量检测的精度、速率以及自动化程度等,基于模型分析的三维检测取代了传统的手工检测和二维检测,成了工业检测领域的重要手段。点云配准作为三维检测中的关键环节,其精度直接影响检测结果的准确性。因此,对国内外学者在点云配准技术方面的主要研究成果进行综述,从算法原理出发,将目前的配准方法归纳为传统配准方法、基于仿生群智能优化算法的配准方法和基于深度学习的配准方法。详细介绍了各类方法的特点、优缺点、典型算法及其变体,总结了点云配准的技术难点并对未来的发展趋势进行了展望。With the development of the manufacturing industry,the required parts are gradually moving towards larger sizes,complex shapes,and high surface processing quality.Moreover,detecting the quality of parts during the processing is an essential step.In order to improve the accuracy,the speed and the automation of quality inspection,the 3D inspection based on model analysis has replaced the traditional manual inspection and the 2D inspection,and becomes an important means in the field of industrial inspection.The accuracy of point cloud registration,as a key part of 3D inspection,directly affects the accuracy of detection results.Therefore,the main research achievements of scholars at home and abroad in point cloud registration technology are summarized.Based on algorithm principles,the current registration methods are summarized into traditional registration methods,registration methods based on affine swarm intelligent optimization algorithms,and registration methods based on deep learning.The characteristics,the advantages and the disadvantages,the typical algorithms and their variants of each method are introduced in detail.The technical difficulties of point cloud registration are summarized and the future development trend is prospected.

关 键 词：三维检测 点云配准 仿生群智能 深度学习 

分 类 号：TG58[金属学及工艺—金属切削加工及机床] TP311[自动化与计算机技术—计算机软件与理论] TP391[自动化与计算机技术—计算机科学与技术]