基于实时视频的车牌识别系统

【作者】 李晶晶；

【导师】 朱清新； 周海根；

【作者基本信息】 电子科技大学 ， 软件工程， 2010， 硕士

【摘要】 车牌识别系统的存在是利用目前现有的计算机先进技术来实现对图像的处理,通过识别技术的应用来实现对车辆号牌的自动认读,提高车辆识别的效率,且对交通流没有影响。车牌识别系统在交通管控、交通流量管理、交通事故自动测报等多个方面均发挥了重要作用,被广泛应用于停车场、港口和机场、城市监控系统、公路流量观测站、桥梁和公路收费站、公路流量观测站等认证车牌的实际交通系统中,论文针对目前车牌识别系统存在的主要问题设计实现基于实时视频的车牌识别系统并对相关算法进行优化。对于定位车牌所存在的区域困难现象,本文通过用车牌的区域定位算法来解决,这个算法是以固定颜色搭配为基础。由于车牌所设定的底色和字符色是由几种确定搭配组成的特性,提取彩色图像,同时以车牌区域的纹理特征进行定位车牌。运用该算法所具有的准确率较高而且所用的时间较短的特点,在复杂或者特殊情况下仍旧能够达到96%的定位率。本文提出一种随机直线检测法来定位车牌的边框,通过这种检测可以解决车牌所具有的倾斜角问题。与采用的Hough算法相比所用的时间较短,工作量也较少。本文在对局部与全局两类阈值法所具有的优劣的基础上,研究得出了图像二值化算法,该算法是局部与全局两类阈值法紧密结合的结果。研究得出的新算法能够对图像的细节进行保留,有效的清除伪影的干扰。图像二值化后车牌上的字符成为单独的字符,通过BP神经网络的识别,字符均识别率95.7%。通过采用以上优化算法,在深入分析车牌识别系统需求基础上设计实现了基于实时视频的车牌识别系统。更多还原

【Abstract】 Based on computer image processing and recognition technology, vehicle recognition system achieves the automatic recognition of license plate reading, vehicle identification greatly improved work efficiency, without increasing the impact on traffic flow. License Plate Recognition System in traffic control, traffic management, traffic accidents and many other aspects of automatic forecasting have played an important role, is widely used in car parks, ports and airports, urban monitoring systems, road traffic observation stations, bridges and roads closed fee station, highway traffic observation station certification number plate, the actual traffic system, the paper license plate recognition system for the current main problems in the design to achieve real-time video-based license plate recognition system and related algorithms optimized.For the difficulties of positioning vehicle plate, this dissertation solves the problem by the location algorithm based on a fixed color matching. As the background color and character color of the plate is determined by several characteristics with the composition of extracted color image. Then, the plate’s region is located by the texture of vehicle plate. This algorithm has a higher accuracy and shorter periods of time, it can still reach 96% of the targeted rate by the characteristics of the complex or exceptional circumstances. This dissertation presents a stochastic line detection method to locate the license plate of the border, and the algorithm can resolve the tilt angle of the license plate. In comparison with Hough method, this algorithm has less calculation and takes shorter periods of time. In this dissertation after analyzing the advantages and disadvantages of global and local algorithms, a new algorithm is put forward which is derived from the combination of them. The new algorithm can keep the details of the image, and effectively remove artifacts possess interference. After image binarization, the characters on the plate split into separate characters, and the characters are recognized by BP neural network, average rate of recognition is 95.7%. By using the above optimization algorithm, a real-time video-based vehicle plate recognition system is achieved in-depth analysis of demand on the vehicle plate recognition system.更多还原