车牌自动识别系统的设计与字符识别算法研究

【作者】 徐国强；

【导师】 解梅；

【作者基本信息】 电子科技大学 ， 信号与信息处理， 2010， 硕士

【摘要】 随着经济的发展,道路交通压力已经成为亟待解决的问题之一,对交通控制、安全管理也提出了更高的要求,这时,运用电子信息技术的车牌识别系统(License Plate Recognition)便应运而生,车牌识别系统的设计,是现代化的智能交通系统得以实现的重要保证。它可以很好地运用在高速公路自动收费系统、大型安全停车场、高速公路车流量统计等场合中。当前,车牌识别很多都是基于PC平台来完成的,当然其具有很容易实现的优势,但是成本高、实时性不强,在稳定性方面表现也比较差,为了更好地实现车牌系统的功能,实现脱机工作,本文所采用的硬件系统为单DSP+FPGA和双DSP+FPGA双板子的方式来共同实现。当然,为了更好地解决系统的识别率和成本问题,车牌识别算法也就成为了系统设计的核心问题之一,硬件系统中,本人完成的主要工作是:(1)对整个硬件系统模块进行熟悉,并完成系统液晶显示模块的调试工作,以使得系统实现可视化;基于CSL对DSP上的底层驱动进行编写。(2)完成车牌识别算法往硬件嵌入式系统中的移植并对其优化,同时完成实验场景下的系统性能测试。软件设计中,在经过系统前端的视频采集装置获取图像后,需要完成对车牌的预处理,之后要很好地定位出车牌区域,接着对其进行车牌字符的分割,以获得后期车牌字符识别的可靠的数据源,在本文中,车牌字符识别算法是关键,为了很好地完成识别算法,我们采用的是将彩色图像转化为二值图像,同时,运用支持向量机(Support Vector Machine, SVM)对分割出来的字符图像进行多分类识别。在对多种核函数在车牌识别中的性能、识别率等综合考虑后,我们提出用径向基核来进行车牌字符识别,训练之前,需要对字符进行必要的预处理,以更好地提高识别率。本文在继承团队成员对车牌字符进行定位和分割之后,完成车牌字符识别算法设计,算法已能够在软件平台上正确实现,识别率良好。更多还原

【Abstract】 With the economic development, modern traffic pressure has become one of the problems which should be given top priority to solve,and also, traffic control, safety management has been put forward higher requirements,At this point, license plate recognition system (License Plate Recognition) which based on electronic information technology has come into being,License Plate Recognition System is an important guarantee for modern intelligent transportation systems to realize. It can be well used in many occasions, such as: highway automation electronic toll collection system, large security parking place, highway traffic volume statistics, and so on.At present, when it comes to license plate recognition, most of them are based on PC platform to work, and obviously, it’s very easy to realize, but as for characteristic of cost, real-time, it’s not feasible, and also, the stability of performance is relatively poor, in order to achieve the license plate system function better, and make it can work offline. In this paper, a single DSP + FPGA and dual-DSP + FPGA dual-board approach are adopted. Of course, in order to solve the system’s recognition rates and costs better, license plate recognition system design algorithm has become one of the core issues. In hardware systems, complete the main tasks which I am responsible for are:(1) To get familiar with the hardware system, and finish the debugging work on the liquid display part; Programming driving section base on CSL for DSP.(2) To transplant license plate recognition algorithm to the embedded systems and to optimize the program, and then, to complete real environment test for system to check its function good or not.In software design part, after obtaining images from system front-end video capture device, something we should do is to complete the pretreatment of the plate, then to properly locate the plate region, and then we should finish license plate character segmentation, in order to obtain reliable data sources for character recognition. In this article, License Plate Recognition is the key. In order to complete the identification algorithm well, we have adopted the way to transform color images into binary images, and to split up the characters in the image on the multi-classification by the use of support vector machine (Support Vector Machine, SVM). By comprehensive consideration and comparison of kernel functions in the vehicle license plate recognition performance, recognition rate, we propose to use RBF kernel to help us complete the license plate character recognition. Before the training, some preprocessing is essential, so as to increase recognition rate better.In this paper, after understanding the work of team members about license location and license plate character segmentation, I finished the design of license plate character recognition algorithm, the algorithm has been able to achieve on software platform correctly, and turned out a good recognition rate.更多还原