摘要
针对立针加工中微孔加工位置定位问题,提出了一种基于模板匹配粗定位和基于边缘检测精定位的两步视觉定位方法。在粗定位阶段,提出了一种基于形状模板匹配和支持向量机的粗定位方法,首先建立了带掩膜的模板,在形状模板匹配的基础上,通过提前终止策略、图像金字塔,以及减少搜索面积的方法对匹配过程进行加速,在对匹配结果提取方向梯度直方图特征后,基于支持向量机进行分类从而防止误匹配。在精定位阶段,提出了一种具有角度约束的边缘检测算法定位外轮廓边缘,并采用二次函数拟合梯度极大值实现亚像素边缘定位,利用最小二乘法拟合轮廓边缘后定位微孔加工位置。在实验阶段,首先通过少量样本对支持向量机进行训练,准确率达到了99.76%,在此基础上本文粗定位方法交并比达到82.9%,而传统的模板匹配方法仅为36.5%;将立针以不同姿态放置在不同位置,在不同光照强度下加入高斯噪声后进行精定位实验,发现本文精定位方法的最大定位误差为2.20μm,而基于Canny边缘检测方法的最大定位误差为5.99μm,本文算法总耗时仅需328 ms。
A two-step visual positioning method based on template matching for rough positioning and edge detection for fine positioning is proposed to solve the problem of locating microholes of bonding wedges.In the step of rough positioning,a rough positioning method based on shape template matching and support vector machine is proposed.Firstly,the template with mask is established,and then it is based on shape template matching that the matching process is accelerated by an early termination strategy,image pyramid and reducing search area.After extracting the histogram features of oriented gradient from matching results,the classification is carried out based on support vector machine to prevent mismatching.In the fine positioning stage,an edge detection algorithm with angle constraints is proposed to accurately locate the edge of the outer contour.The sub-pixel edge positioning is realized by using a quadratic function to fit the maximum gradient,and the least square method is used to the contour edge to locate the machining location of microholes.In the experimental stage,the SVM is trained with a small number of samples,and the accuracy reaches 99.76%.On this basis,the intersection of union of the proposed coarse positioning method reaches 82.9%,while that of the traditional template matching method is only 36.5%.The fine positioning experiment is carried out by placing the bonding wedges in different positions with different postures and adding Gaussian noise under different light intensities.The maximum positioning error of the proposed fine positioning method is 2.20 microns,while the maximum positioning error based on Canny edge detection method is 5.99 microns.The proposed whole algorithm only takes 328 ms.
作者
毕逸飞
姜歌东
梅雪松
黄旺旺
王高才
BI Yifei;JIANG Gedong;MEI Xuesong;HUANG Wangwang;WANG Gaocai(School of Mechanical Engineering, Xi’an Jiaotong University, Xi’an 710049, China)
出处
《西安交通大学学报》
EI
CAS
CSCD
北大核心
2021年第8期59-67,共9页
Journal of Xi'an Jiaotong University
基金
国家重点研发计划资助项目(2017YFB1104602)。
关键词
芯片封装
立针
微平面
微孔
模板匹配
边缘检测
chip packaging
bonding wedges
micro-plane
microhole
template matching
edge detection
