车牌识别算法及其在QT平台上的实现

发布时间：2018-05-08 06:50

本文选题：图像能量 + 车牌定位　；参考：《北方工业大学》2017年硕士论文

【摘要】：由于机动车数量的急剧增长,普通的交通管理方式无法满足对现有车辆的管理和流量调控的需求。智能交通系统概念的提出是为了建立更加智能化、快捷化和自动化的交通系统,以缓解交通的拥堵的问题。而智能交通的核心的技术之一就是车牌识别技术,车牌识别技术的研究对于建立智能交通系统提供了坚定的技术基础。车牌区域的定位和车牌字符的识别作为车牌识别系统的两个核心部分,决定着整个系统的性能。本文通过对车牌识别系统的文献的大量阅读和研究后,通过综合的学习和比较,设计并实现了基于图像能量和神经网络的车牌识别系统,并采用Qt平台完成了系统简易界面的设计。通过对常用车牌区域定位算法的研究和总结,实现了基于图像能量的车牌定位算法,并在试验中进行了验证。该算法首先构造预处理后的灰度车牌图像的能量图像,然后初步定位出能量图像中能量最大的三个候选区域,最终筛选和验证初步定位出的候选区域,完成车牌区域的准确定位并得到车牌区域。通过对常用的车牌字符识别算法的深入研究和对比后,采用二级BP神经网络的方法作为车牌识别系统的字符识别方法。该算法改进了字符特征的提取方法,并采取了二级的BP神经网络:第一级包括字符网络和数字字母网络,第二级是相似字符网络。大量实验证明,通过构造相似字符网络可以很大程度的提高字符识别率。本系统基于Qt开发平台设计车牌识别系统的GUI界面,具有一定的可操作性及对不同的硬件平台的跨平台特性。实验结果表明,本车牌识别软件系统可以识别不同角度,不同距离和不同光照环境条件下拍摄到的车牌图像。由于引入Qt,方便软件分别在Win+PC机或Linux+ARM系统上运行,具较强的应用前景。
[Abstract]:Because of the rapid increase of the number of motor vehicles, the common traffic management mode can not meet the demand of the existing vehicle management and flow control. The concept of Intelligent Transportation system (its) is proposed in order to set up a more intelligent, fast and automatic traffic system to alleviate the problem of traffic congestion. One of the core technologies of intelligent transportation is license plate recognition. The research of license plate recognition provides a firm technical foundation for the establishment of intelligent transportation system. The location of license plate region and the recognition of license plate characters are the two core parts of the license plate recognition system, which determines the performance of the whole system. After reading and studying a lot of documents of license plate recognition system, this paper designs and implements a license plate recognition system based on image energy and neural network through comprehensive learning and comparison. QT platform is used to design the simple interface of the system. Based on the research and summary of common license plate location algorithms, a license plate location algorithm based on image energy is implemented and verified in the experiment. The algorithm firstly constructs the energy image of the pre-processed gray license plate image, and then preliminarily locates the three candidate regions with the largest energy in the energy image, and finally selects and verifies the candidate regions that are initially located. Complete the accurate location of the license plate area and get the license plate area. After the deep research and comparison of the commonly used license plate character recognition algorithms, the two-stage BP neural network method is adopted as the character recognition method of the license plate recognition system. The algorithm improves the method of character feature extraction, and adopts the two-level BP neural network: the first level includes character network and numeric alphabetical network, the second level is similar character network. A large number of experiments show that the character recognition rate can be greatly improved by constructing a similar character network. The GUI interface of license plate recognition system is designed based on QT development platform. It has some maneuverability and cross-platform characteristics for different hardware platforms. The experimental results show that the license plate recognition software system can recognize the license plate images under different angles, distances and illumination conditions. Because of the introduction of QT, it is convenient for software to run on Win PC or Linux ARM system, which has a strong application prospect.
【学位授予单位】：北方工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】