【作者】 周燕；

【导师】 段其昌；

【作者基本信息】 重庆大学 ， 控制理论与控制工程， 2008， 硕士

【摘要】 智能运输系统是21世纪现代化交通运输体系的重要发展方向,它是一种信息化、智能化、社会化的新型现代交通系统。随着社会经济的不断发展和人们生活水平的日益提高,整个社会对交通运输的需求不断增大,智能交通系统的应用势在必行。实时车牌识别作为智能交通系统中的一个分支,在大型停车场的管理系统、公共安全、交通管理及有关部门有着特别重要的实际运用价值,正日益受到人们的重视。本文运用图像处理技术、模式识别技术、车牌定位技术、字符分割技术、神经网络识别技术等来解决车辆牌照识别问题;分为图像预处理、车牌区域定位与几何校正、字符分割与归一化及字符识别四个模块对实时车牌识别进行分析研究。论文首先研究图像的预处理,将采集到的含有车牌的图像进行灰度化、灰度拉伸、直方图均衡化、中值滤波和图像二值化处理,并改进了一种二值化效果较好的基于灰度直方图的全局最佳平均阈值法;然后对我国车牌特征和常见的车牌定位技术进行深入研究,改进了一种基于车牌区域灰度分布与几何特征的实时车牌定位方法;通过对车牌发生几何形变的原因及类型的分析研究,采用了基于斜率的车牌倾斜校正方法;基于常见的字符分割方法和车牌本身的结构特点及先验知识的分析研究,改进一种基于投影法与车牌先验知识相结合的分割方法;采用邻近插值法,虽然精确度相对较低但可以满足系统的要求且实现方便,将字符归一化为32×16的点阵,为字符的识别做好准备。最后对粒子群优化算法和神经网络的相关原理作了简要介绍和分析研究,在此基础上提出了PSO-BP神经网络并将其用于实时车牌字符识别。更多还原

【Abstract】 Intelligent Transport System in the 21st century modernization of the transportation system an important development direction, it is a kind of information, intelligence, and a new type of modern society of the transport system. With the continued socio-economic development and increasing the living standards of the people, the entire community of transport demand growing, Intelligent Transportation System Application inevitable. Real-Time License Plate Recognition as a branch of Intelligent Transportation System ,in the large car park management system, public safety, traffic management is particularly important and relevant departments have the practical application of value, are increasingly subject to the people’s attention.This paper uses image processing technology, pattern recognition technology, the vehicle registration positioning technology, character segmentation techniques and neural network technology to solve the problem of vehicle licence identification. Divided into image preprocessing, regional location and license plate geometric correction, character segmentation and normalized and Character Recognition of the four modules for real-time analysis of License Plate Recognition.First papers on image preprocessing, will be collected on the plate containing the image of the gray, gray stretch, histogram equalization, filtering and image binarization handling, and a binarization based on the good results of the histogram best overall average threshold value; Then on China’s plate features and common license plates in-depth study of positioning technology, a license plate based on regional distribution and gray geometric characteristics of the real-time positioning methods plate through the plate geometry deformation occurred and the reasons for the type analysis of the slope is based on the license plate tilt correction method; Based on the common character segmentation methods and the structural characteristics of its own license plate and a priori knowledge of the analysis, a projection based on a priori knowledge of the plate and the combination of segmentation method used to neighbouring interpolation method, although the relative accuracy But to meet the low system requirements and easy, characters will be normalized into the lattice, character recognition prepared. Finally, the PSO algorithm and the related principles of neural networks in brief and analytical studies, based on this highlights the PSO-BP neural network for real-time and its license plate character recognition.更多还原