基于机器学习的车牌识别算法研究

发布时间：2018-03-09 02:22

本文选题：人工神经网络　切入点：机器学习　出处：《沈阳师范大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着智慧城市战略在国家重要城市快速的推进与发展,城市中各项运行的核心系统都将通过运用新一代的技术手段进行分析、整合、处理。公共交通信息,作为城市日常生活中必不可少的一员,在很大程度上需要快速高效的进行管理与运行。本文通过对一个公共交通信息管理系统拓展功能需求进行研究,利用机器学习中的技术对系统中包含车牌信息的图像进行处理,并在车牌智能识别方面上对于国内外识别算法文献进行了分析与讨论,综合考虑本系统特点,实现了本文车牌字符自动识别的过程。本文车牌自动识别算法研究的理念主要来源于开源系统EasyPR,在其基础上结合本系统需求进行分析与改进,本文车牌识别模块算法基于OpenCV(Open Source Computer Vi sion Library,即OpenCV)跨平台计算机视觉库,轻量、高效。车牌识别过程主要分为两大部分:1)车牌图块定位:通过对分析和处理一张包含车牌的图片,得到预选车牌图块,再对预选车牌图块进行判断,最终得到只包含车牌信息的图块。本阶段定位方法采用颜色空间HSV模型与边缘检测算法Sobel算子综合定位,车牌判断过程主要通过基于支持向量机(S upport Vector Machines,简称SVM)算法训练、学习,并判断得到的图块是否为真正的车牌。2)车牌字符识别:对通过定位得到的车牌图块进行字符分割、字符识别,然后顺序排列输出字符型车牌信息。本阶段字符分割主要根据边缘轮廓提取、外接矩形等方法,字符识别过程主要通过人工神经网络(Artificial Neural Network,即ANN)种两层神经网络模型进行学习和分类输出。本文车牌字符识别模块主要算法基于OpenCV开源视觉库,通过C++进行编写,采用VS2013进行运行与测试。本文采用的车牌自动识别算法在实验结果中显示具有一定的定位准确率与识别率,基本符合系统所需需求的多项条件。
[Abstract]:With the rapid promotion and development of the smart city strategy in the important cities of the country, the core systems that run in the cities will be analyzed, integrated and processed through the use of the new generation of technical means. As an indispensable member of urban daily life, it needs to be managed and operated quickly and efficiently to a great extent. In this paper, the functional requirements of a public transportation information management system are studied. This paper uses the technology of machine learning to process the image which contains the license plate information in the system, and analyzes and discusses the domestic and foreign literatures on the recognition algorithm of the license plate in the aspect of the intelligent recognition of the license plate, considering the characteristics of the system synthetically. The research idea of automatic recognition algorithm of license plate mainly comes from the open source system EasyPR.Based on the analysis and improvement of this system, The algorithm of license plate recognition module is based on OpenCV(Open Source Computer Vi sion Library, which is light and efficient. The process of license plate recognition is mainly divided into two parts: 1) location of license plate block by analyzing and processing a picture containing license plate. The pre-selected license plate block is obtained, and then the pre-selected license plate block is judged. Finally, the block containing only license plate information is obtained. In this stage, the color space HSV model and the edge detection algorithm Sobel operator are used to locate the image. The process of license plate identification is mainly trained based on support vector machine (SVM) S upport Vector machines (SVM) algorithm, learning, and judging whether the obtained block is a real license plate .2.) license plate character recognition: the character segmentation of the license plate block obtained by location is carried out. Character recognition, then order the output of character license plate information. This stage of character segmentation is mainly based on edge contour extraction, external rectangle and other methods, The character recognition process is mainly carried out through artificial Neural Network (Ann) two-layer neural network model. The main algorithm of the license plate character recognition module is based on OpenCV open source vision library, and is written by C, the main algorithm of this paper is based on the open source vision library of OpenCV. The automatic license plate recognition algorithm adopted in this paper has a certain accuracy and recognition rate in the experimental results, which basically meets the requirements of the system.
【学位授予单位】：沈阳师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TP181

【参考文献】