【作者】 廖晓姣；

【导师】 李英；

【作者基本信息】 江南大学 ， 检测技术与自动化装置， 2012， 硕士

【摘要】 车牌识别系统作为智能交通系统的重要组成部分,它在违章抓拍、停车场管理、智能小区交通管理和重要关卡车辆登记等方面有重要作用。本文从实际应用角度出发,针对城市道路以及高速公路等室外背景,对车牌识别系统中车牌定位、字符分割、字符识别等算法系统地进行了研究,论文主要工作如下：在对现有的车牌定位算法进行研究比较的基础上,采用基于边缘检测和形态学处理的车牌定位算法,对Sobel边缘检测进行了改进,通过设定阈值来降低周围背景中细小边缘的干扰；并利用基于形态学处理的连通域分析,完成车牌定位。在字符分割方面,采用基于最小二乘和最小投影距离相结合的车牌倾斜校正法方,在水平校正时,利用最小二乘拟合直线的方法求取倾斜角；在垂直校正时,利用了改进的最小投影距离法,降低了算法的执行次数,提高了执行效率。对校正后的车牌又利用投影法进行分割,并对存在粘连和分裂情况的字符进行了特殊处理。在特征提取方面,采用基于小波变换和Zernike矩的组合特征,小波变换系数提取了字符的细节特征,Zernike矩提取了字符的全局特征,二者的组合特征更全面的区分了每个字符,并利用主元分析法对小波变换系数进行降维,去除相关性较大的量,降低了算法的运算量。最后对神经网络字符识别和支持向量机字符识别进行了分析,利用支持向量机设计车牌字符分类器,不仅识别率高、测试误差小,而且具有较强的泛化能力。最后,在Matlab仿真平台上,对大量车牌图像进行了测试,结果表明,所改进的算法具有较高的有效性和实际应用价值。更多还原

【Abstract】 As an important part of intelligent transportation systems, the license plate recognition system has an important role in violation enforcement, parking management, intelligent traffic management and the important points of vehicle registration. From the view of practical application, some researches have been done on license plate location, license plate segmentation and character recognition against the outdoor background of urban roads and highways. The main work is as follows:On the basis of studying the existing vehicle license plate location algorithms, An License plate location based on edge detection and mathematical morphology has been presented.The Sobel edge detection has been improved, by setting the threshold to reduce the small edges’disturbance around the background, through using connected domain analyses based on morphological processing to complete the license plate location.In terms of character segmentation, first, a tilt correction of license plate based on least squares and minimum projection distance has been put forwarded, on level correction, using least square to fit a straight line for obtaining the slope angle, on vertical correction, using an improved minimum projection distance method, it has reduced the number of execution and improved the execution efficiency. To the licenses which have been corrected has been segmented using modified projection method. Also the characters which are adhesion and fragmentation have been given special treatment.In terms of extraction, a feature extraction based on Wavelet transform and Zernike has been presented.wavelet transform has extract the details of character features, Zernike moments has extract the global character features, the combination of those has distincted the character between each other. And using principal component analysis to reduce some features which have large correlation with others, the number of the operation of algorithm also has been reduced.Last, based on the analysis of the neural networks and support vector machine(SVM), using SVM to design the network of license plate characters, it has not only high recognition rate, small measurement error, but also a strong generalization.Finally, on the Matlab simulation platform, a large number of plate images has been tested. Results show that the improvement algorithms have high effectiveness and the value of practical application.更多还原