【作者】 朱浩；

【导师】 胡峰松；

【作者基本信息】 湖南大学 ， 计算机科学与技术， 2015， 硕士

【摘要】 随着人们生活水平的不断提高以及科技的不断进步,汽车已深入各家各户,成为人们出行的必备工具之一,进而使得道路上机动车辆数目不断增加,交通拥挤、堵塞、违章等问题也越来越多,如何实现对车辆进行有效地管理便成了智能交通领域中的重要研究课题。基于数字图像处理以及模式识别的车牌识别技术主要是将汽车牌照从复杂背景中提取并识别,通过预处理、车牌定位、字符分割以及字符识别等技术,识别出车牌号码,为智能交通系统中实现车辆管理提供了强有力的保障。当前车牌识别技术已经应用到公路收费、停车场与小区车辆管理、城市道路交通管理等场合,具有广阔的应用前景,而车牌定位作为车牌识别技术中必不可少的模块,也具有重要的研究意义与价值。本文对车牌识别技术的基本原理以及基本模块进行详细介绍,论述车牌区域的基本特征,分析目前常用的国内外车牌定位算法的基本原理以及优缺点,并详细阐述我国车牌定位的难点所在。此外还对图像预处理技术以及图像局部特征提取与匹配SURF算法的原理、特征提取与匹配过程进行了分析与介绍。论文的主要研究工作如下:(1)充分利用车牌区域的几何、纹理以及颜色特征,提出了一种基于HSI颜色空间和行扫描的车牌定位算法。首先利用车牌的颜色对彩色车辆图像进行颜色分割,在结合车牌的面积、长宽比等几何特征进行筛选,得到车牌候选区域,再结合车牌的纹理特征,利用行扫描算法分析候选区域每行的连续跳变数目,确定车牌的上下边界,最后利用垂直投影法获取车牌的左右边界,精确定位车牌区域。通过实验对比分析,该算法具有较好的定位效果。(2)分析了复杂背景下利用SIFT算法进行车牌精确定位的不足,同时结合SURF算法在图像特征提取与匹配方面的优点,充分利用车牌的多重特征,提出了一种基于多重特征和SURF算法的车牌提取方法。该算法首先根据车牌的颜色、几何以及纹理特征筛选获取车牌候选区域,再利用SURF算法提取车牌标准汉字模板库图像与候选区域进行特征匹配,精确定位车牌区域。通过实验对比分析,该算法在时间与准确率方面均具有不错的效果,而且还能实现对车牌首个汉字字符的识别,减轻后续字符识别的负担。更多还原

【Abstract】 With the continuous improvement of people’s living standards and the advances in technology,automobile has penetrated each household and it has become an essential tool for people to travel.As the increasing number of motor vehicles on the road,traffic congestion,congestion,illegal and other issues become more and more,how to effectively manage the vehicle has become an important field of intelligent transportation research.The license plate recognition technology based on digital image processing and pattern recognition mainly extract license plate from complex background and identify it.Through preprocessing,license plate location,character segmentation and character recognition technology,it can identify the license plate number and provide a strong guarantee for intelligent transportation systems achieve vehicle management.Today license plate recognition technology has been applied to the road toll,parking and vehicle management district,urban road traffic management and other occasions,and it has broad application prospects.The license plate location is an essential module in plate recognition technology,it also has important significance and value.The paper describes the basic principle of the license plate recognition technology and basic modules in detail,discusses the basic features of the license plate area,analysis the basic principles,advantages and disadvantages of the current domestic and foreign common license plate location algorithms,and elaborates the difficulties in locating the license plate.In addition,the image processing technology and the principle,feature extraction and matching process of local image feature extraction and matching SURF algorithm are analyzed and described.The main research work are as follows:(1)Making full use of geometry,texture and color characteristics of the license plate area,this paper proposes a license plate location algorithm which is based on HSI color space and line scanning.Firstly,it uses the vehicle license plate color to segment the image,combines with the area,ration of geometric features in the license to obtain a license plate candidate region.And then combined with the texture plates,it use line scanning algorithm to analysis the number of consecutive jumps of each row in candidate regions to determine the license plate of the upper and lower boundaries.In order to get the left and right borders of the license plate,it use verticalprojection algorithm.Through comparative analysis of the experiment,the algorithm has better positioning effect.(2)Through analysis the deficiencies of using SIFT algorithm to locate the license plate in complex background,and combined with the advantages of image feature extraction and matching SURF algorithm and the full use of multi-feature license plate,it proposes a license plate location method which is based on multi-feature and SURF algorithm.First,it uses the colors,geometric and texture features to get the license plate candidate region,and then it use SURF algorithm to extract the license plate characters standard template library and the candidate region image to match feature,positioning plate area precisely.Finally,through the comparative analysis of the experiment,the algorithm not only has good results in time and accurate rate,but also it can achieve the recognition of the license plate of the first Chinese character,reducing the burden on the subsequent character recognition.更多还原