【作者】 张凯；

【导师】 王瑜辉； 熊有伦；

【作者基本信息】 华中科技大学 ， 机械电子工程， 2005， 硕士

【摘要】 随着电子元器件集成度的增加和尺寸的减小,以及生产效率的逐年提高,电子制造设备的运动精度和速度要求变得越来越高。视觉检测和定位技术作为保证电子制造设备高速、高精运行的核心关键技术,成为各大设备制造厂商和学术界的研究热点。本文对电子制造设备视觉系统前端的成像光路设计和后端的视觉处理算法进行了研究。针对电子制造中对芯片进行高速封装的需求,本文提出一种新的飞行视觉方案,可有效减少视觉校正的时间。该方案在分析以前专利文献中出现的静止反射镜法等方案优缺点的基础上,使用一个可移动的反射镜,当反射镜处于不同的工作位置时分别获取元件和工件表面的图像。论文对该方案的实现细节进行了讨论,设计了一种斜视摄像机的位姿微调机构,给出了透视失真图像的校正方法。在视觉处理算法的研究方面,本文搭建起一个集成数字摄像机和图像处理软件的视觉处理算法实验平台,实现了灰度图像的获取、显示、存储及各种常用的图像处理功能。论文对电子制造中的视觉检测内容及其算法进行了归纳,并以三个具体的视觉检测应用为例,说明了在平台上进行的典型算法实验研究工作。结合在平台上进行算法优化实验的经验,提出了视觉处理算法的五条设计原则。最后,论文针对电子制造设备中常见定位目标的对准,提出了一种新的识别定位算法,即通过改进一种混合空间增强的预处理方法,获取其边缘几何特征为识别定位对象,通过特征变换减少特征匹配的数量,这种基于几何特征的算法比现有的识别定位方法具有更快的定位速度,同时对光照变化、比例变化、局部特征丢损等工程实际现象的处理有较强的鲁棒性。更多还原

【Abstract】 The electronic manufacturing devices are required to have better performance in speed and accuracy to meet the need of high density packaging, component size minimizing and high speed assembling. The visual inspection and alignment technology play a key role in achieving the high performance. In this dissertation, the problem of optics path design, optics inspection algorithm optimization and vision alignment algorithm development is addressed. A new on-the-fly vision system is designed to meet the requirements of high speed mounting of the components. The former on-the-fly vision systems are first analyzed and the virtues and shortcomings are explored. Unlike the stationary mirror method used in former designs, a retractable mirror is used to get the image of the component and the workpiece during the movements of the pick-up head. The implementation issues are further discussed. A fine adjustment mechanism is designed to correct the location and orientation of the camera, and a perspective distortion correction method is also proposed. A test framework for the vision processing algorithm is developed. The framework is designed with the ability of gray scale image acquisition, displaying, saving and processing. The optical inspection task and the corresponding algorithms in the electronic manufacturing are analyzed. The experiment and research work carried out in the framework are illustrated by 3 examples of real inspection tasks. On the experience of algorithms optimization, five principles are proposed for the design of the vision algorithms. A fast geometric characteristic-oriented algorithm is proposed for the cross fiducial alignment to meet the critical requirement of fast and accurate chip mounting. A hybrid spatial filtering method is improved to get a high quality bi-level image, then the randomized Hough transform and the least square method is taken to calculate accurate border line parameters. Experiments results demonstrate that the algorithm is fast and robust to the illumination changes, scale changes and partially hidden of the fiducial mark.更多还原