基于Hu-SIFT特征匹配与遗传算法的零件视觉分拣方法  

Visual sorting method of mechanical parts based on Hu-SIFT feature matching and genetic algorithm

作　　者：鲁晟燚 梁冬泰[1] 梁丹 吴晓成 LU Shengyi;LIANG Dongtai;LIANG Dan;WU Xiaocheng(Faculty of Mechanical Engineering and Mechanics,Ningbo University,Ningbo 315211,China)

机构地区：[1]宁波大学机械工程与力学学院,浙江宁波315211

出　　处：《传感器与微系统》2020年第12期123-126,共4页Transducer and Microsystem Technologies

基　　金：国家自然科学基金青年科学基金资助项目(51805280);宁波市自然科学基金资助项目(2019A610158);宁波市科技创新2025重大专项项目(2018B10005)。

摘　　要：针对机械零部件快速分拣需求,提出一种基于Hu-尺度不变特征变换(SIFT)特征匹配与遗传算法的零件视觉分拣方法。首先,通过形态学处理与滤波算法对零件图像进行预处理;然后,设计基于Hu-SIFT特征匹配的图像识别算法,融合Hu不变矩和SIFT特征进行全局和局部的特征匹配,以准确提取目标零件,并利用仿射变换进行目标零件定位;最后,针对传统遗传算法收敛速度慢的问题,提出一种基于极值优化的改进遗传算法,对机械手目标零件的最优分拣路径进行快速规划。实验表明:所提方法的平均查准率和查全率为86.06%和93.25%,最优路径规划的平均时间为3.03 s,可有效提高机械手的分拣效率。For the needs of rapid sorting of mechanical parts,a visual sorting method of mechanical parts based on Hu-scale invariant feature transform(SIFT)feature matching and genetic algorithm is proposed.Firstly,the part image is pre-processed by morphological processing and filtering algorithm.Secondly,an image recognition algorithm based on Hu-SIFT feature matching is designed.Hu Invariant moment and SIFT are fused to match global and local features,to extract target parts accurately.And affine transformation is used to locate the target parts.Finally,focusing on the problem of slow convergence of traditional genetic algorithms,an improved genetic algorithm based on extremal optimization is proposed,planning the optimal sorting path of manipulator target parts.The experiment results show that the average precision and recall rates of this method are 86.06%and 93.25%.The average time of optimal path planning is 3.03 s,which can effectively improve the sorting efficiency of the manipulator.

关 键 词：零件分拣 特征匹配 机器视觉 路径规划 

分 类 号：TP249[自动化与计算机技术—检测技术与自动化装置] TP212[自动化与计算机技术—控制科学与工程]