跟踪印花系统视觉检测算法研究-模式识别与智能系统专业论文.docxVIP

I I 摘 要 在传统的印花系统中，国内许多纺织企业是通过印花人员目测的方法去检验印 花过程的准确性。该方法存在着标准不一、检测精度不高、劳动强度大等缺点。近 年来，随着机器视觉技术的不断发展与成熟，将计算机图形学应用于工业印花系统 成为了可能，为了研制领先于行业的跟踪印花系统，本文主要针对系统的算法部分 做了较为深刻地探索与研究。 本文首先介绍了在实验室条件下搭建的实验平台，根据跟踪印花系统整体性能 的需要，给出了一套切实可行的硬件解决方案，并对硬件模块的几个重要组成部分 （CCD 相机、光源、图像采集卡）进行了详细的阐述。 在跟踪印花系统中，核心算法的好坏直接决定着系统的成败。检测算法作为最 为关键的模块，既要保证检测精度同时又要满足系统的实时性约束，因此系统分别 从横向、纵向两个方向去实现跟踪印花系统的算法实现。在布匹运行方向，针对印 花布匹的特殊性，本文提出了基于直方图以及投影的特征的模板匹配算法，首先通 过比较直方图特征进行粗匹配，挑选出模板存在可能性较大的区域，然后通过计算 横向与纵向的投影特征来实现模板的精确匹配；而在布匹的纬度方向，由于可供选 择的图像模板比较大，从而提出了基于豪斯多夫距离的检测算法，首先利用边缘检 测的方法提取目标图案的大致轮廓，然后通过单向豪斯多夫距离进行模板的精确匹 配。为了进一步优化算法，本文还提出了自适应阈值分割的方法。大量的实验结果 显示，与当前已经存在的众多图像算法相比，本文提出的算法计算量大幅减少，既 可以保证得到高精度的检测结果，同时能够较好地满足系统实时性约束。 关键词： 机器视觉，模板匹配，边缘检测，豪斯多夫距离 II II Abstract In traditional printing system, many domestic factories detect the accuracy of the printing process by human vision. This method has many shortcomings, such as lack of uniform standard, low detection accuracy, high labor strength and so on. Recent years, along with the development and maturity of machine vision technology, it is possible to use the computer graphics in industrial printing system. In order to develop industry- leading tracking printing system, this thesis mainly studies the algorithmic parts of the system. In this thesis, the structure under laboratory conditions is firstly investigated. Then based on the overall performance of the tracking printing system, a set of practical hardware solution is proposed, and also several important components, such as CCD camera, light source, frame grabber, are described in detail. In tracking printing system, whether the core algorithm is good or bad directly determines the success or failure of the entire system. As the most important part, the detection algorithm should not only ensure the detection accuracy, but also meet the real-time constraint. So the system achieves the tracking algorithm in both horizontal and vertical directions. In the direction of the fabric moving forward, as the printing fabric i