机动车车牌自动识别系统的算法研究

【作者】 苏俊人；

【导师】 马争；

【作者基本信息】 电子科技大学 ， 通信与信息系统， 2010， 硕士

【摘要】 机动车车牌识别(Vehicle License Plate Recognition)系统是智能交通系统的一个重要组成部分,在缓解和管理日益拥堵的城市交通和公路交通中有着举足轻重的作用。目前车牌自动识别系统不仅应用在停车场、高速公路出入口收费、小区车辆管理等简单卡口场景中,也应用在了路况繁杂、交通拥堵的城市十字路口等复杂场景中。本论文根据数字图像处理、模式识别技术以及统计学习理论,主要针对简单卡口和十字路口两种场景提出车牌自动识别系统中相应的车牌定位和字符识别算法。论文的主要研究工作如下:1.车牌定位,即从包含车牌的车辆图像中提取出感兴趣的号牌。本文提出了基于投影和密度的车牌定位方法。在对车辆图像进行灰度化、灰度拉伸等一些预处理后,计算水平区域的一阶差分并进行投影得到其一阶差分图,再经过波峰合并和中值滤波定位出车牌的水平位置;然后利用Sobel垂直算子进行边缘检测,通过密度法定位出车牌的左右边界。最后采用Hough变换对定位后的倾斜车牌进行矫正,为字符分割提供更好的数据源。2.车牌字符识别,即从分割出来的字符图像识别出对应的字符,并以文本形式输出。本文针对两种不同场景分别采用不同的字符识别算法,即针对卡口场景的One-against-One多分类支持向量机算法以及针对十字路口场景的基于Adaboost与多分类支持向量机的组合分类算法。并根据现有车牌标准GA36-2007和GA36-1992结合车牌字符特征的先验知识,将车牌字符分为四大类分别训练,来进一步提高算法的识别率。本文提出的所有算法都经过Matlab7.0仿真验证,并用C语言实现。实验证明针对卡口场景和十字路口场景的车牌定位算法的准确率分别是97.5%和95.9%,字符识别准确率分别是97.65%和91.48%。更多还原

【Abstract】 Automatic Vehicle License Plate Recognition System is an important component of the Intelligent Transportation System. It plays an important role in mitigating and managing the increasing congestion in urban traffic and road transport. At present, the Automatic Vehicle License Plate Recognition System is not only used in car parks, highway tolls, residential vehicle management and other simple gate scenes, and also is used in some complex scenes such as urban crossroads which has a complex road conditions and a heavy traffic.In this paper, according to Digital Image Processing, Pattern Recognition and Statistical Learning Theory, the algorithms of license plate location and character recognition for simple gate scenes and urban crossroads in automatic Vehicle License Plate Recognition System are proposed. Major researches in this paper are as follows:1. Vehicle license plate location is to extract the license plate from which the vehicles images contain. In this paper, an algorithm for the license plate location based on the density and projection is put forward. After the image preprocessing, such as transformation from color images to gray images and gray stretch, calculate the first-order differential of the whole level region to get the first-order image, and then, the level position of plate can be located through the algorithm of wave merging and the median filtering. Next, using the vertical Sobel edge detection operator, the left and right boundary of plate can be located through the density. Finally, using the Hough transform to correct the located inclined plate for a better data source for character segmentation.2. License plate character recognition is to identify the corresponding characters from the character images which are from the character segmentation, and give the results in text form. In this paper, two algorithms of character recognition for two different scenes are used respectively, that is the One-against-One multi-class support vector machine algorithm for the gate scenes and the Adaboost algorithm combined with the One-against-One multi-class SVM for the crossroads. And in accordance with standards of China’s existing license plate GA36-2007 and GA36-1992, combining the features of license plate characters, the license characters are divided into four classes, and trained respectively for a better recognition rate.All algorithms proposed in this paper have been simulated by MATLAB 7.0, and programmed with C. The results show that the accuracy rate of location algorithm for the gate scenes and the crossroads is 97.5% and 95.9%, the accuracy rate of character recognition is 97.65% and 91.48%.更多还原