摘要
车牌识别广泛用于电子收费、出入控制、交通监控等重要场合。车牌定位作为车牌识别中的重要环节,对系统识别精度有重要的影响。目前的车牌定位方法主要是针对所监视的区域只有单一车辆的情况。但在许多情况下,监视区域比较复杂。比如车载监控系统、多车道公路路口的监视与监控、城市要道的监视与监控,所监控的区域一般同时会出现多辆汽车,背景也比较复杂。所以多车辆图像的牌照识别开发就具有重要意义。
基于以上分析,本文着重进行了多车牌定位的研究,提出了一种综合利用边缘检测、连通域分析、倾斜矫正等多种方法相结合的多车牌定位的算法。主要工作如下:
1、对要处理的图像进行预处理,包括彩色图像转化为灰度图像、灰度均衡、小波去噪等,首先彩色图像转化为灰度图像采用现行标准的平均值法,g表示灰度化后灰度值,R,G,B分别表示原彩色图中的红、绿、蓝分量,有g=0.3R+0.59G+0.11B;然后通过对灰度化的图像进行直方图均衡化,消除了光照引起的图像差异,使得图像上明暗对比显著,牌照区域的笔画特征明显;最后对于图像去噪,本文采用了一种小波局部阈值的设置方法来进行图像的去噪,完成图像预处理。
2、对预处理后的图像利用数学形态学运算将含有多车牌的图像中各疑似单车牌的联通域定位出来。首先进行图像边缘检测,由于图像边缘是多样的,为了达到提取图像的多种类型的边缘,本文采用将形态学运算与集合运算结合起来,选取膨胀腐蚀型结构算子,采用多结构元素,即先对小波去噪后的图像分别进行膨胀、腐蚀运算,然后取膨胀后的图像与腐蚀后的图像之差得到较好的图像边缘。其次对边缘检测后的图像进行膨胀运算,填充图像中的“小洞”;接下来进行水平闭运算填平小湖(即小孔),弥合小裂缝,和水平开运算,去除孤立的小点,毛刺和小桥(即连通两块区域的小点);确定出含有车牌区域的多个连通域;最后对经过多种形态运算后的图像进行面积剔除,去除较小面积的连通域,从而几个可能的车牌连通域被确定出来。
3、提出一种改进的Hough变换法进行疑似车牌连通域的纠正。首先通过检测疑似车牌连通域边框直线来获得车牌的倾斜角度,计算出牌照图像的倾斜角度后,以车牌照图像中心位置作为旋转中心来进行旋转,使得各疑似车牌连通域得到旋转校正,旋转后牌照图像能保持水平放置。
4、利用车牌的特征信息对各疑似单车牌区进行了去伪验证,完成车牌准确定位并对实验结果进行了比较。本算法复杂度低,能够满足快速、准确定位要求;并且对背景复杂、光线不均匀、字符和底色对比度低以及车辆分布情况的复杂性等并不很敏感,具有较好的鲁棒性。
The vehicle license plate recognition (LPR) system has a wide application in the automatic tolling system, access control system and traffic monitoring and so on. As an important part of the vehicle license plate recognition, vehicle license plate division (LPD) plays a vital role in the precision of such systems. Present method of LPD focuses mainly on images with single car. Multiple cars appearing in the same image is more common in the real world. What’s more, the vehicle image is most likely with complex scenes as well. To deal with vehicle license plates division in such situations, images with multiple cars in complex scenes are studied in this paper. The image processing technology proposed in this paper can divide the area in which the car plate is located as close as possible. The experimental results show that multiple car plates within the same image in complex scenes can be separated correctly and accurately.
In this paper, current research status and application of LPR are introduced, the developing and defect about LPD are discussed first. Then the key techniques including edge detection, component connectivity analysis, slant correction etc. manifold methods, multiple vehicle license plates location and division are expatiated. The main contributions of this paper are summarized as follows:
1. Several pre-processing methods of image processing are adopted. Including conversion from colored-images into gray-scale images, gray balance and eliminate the differences between images caused by illumination. The pre-processing makes the contrast of images more significant, so the stroke characteristics of license number in the region stand out. The wavelet transform using threshold settings partial approach is introduced to de-noise the image as the last step in pre-processing stage.
2. After the pre-processing stage, mathematical morphology method is used to deal with the images which containing several license plates. The suspected locations of license plate are isolated into different simply connected domain. In order to extract of images which contain various type of edges, the morphology of computing is combined with set computing in this paper. And multiple structure elements of morphology are used to detect the edges of connected domain. During our experiment we found that the result we get by using multiple structure elements of the image edge detection operator of morphological during the dilation and erosion is much better than that we get by using the traditional edge detection methods. After the edge detection, a variety of morphological image processing operations are used to determine the connectivity of the suspected domains.
3. Then, in order to detect the suspected license plates more accurately, a reformed Hough transform was proposed in this paper to correct the tilt of connected domain.
4. Finally by using of license plate information of the characteristics of a single plate of the suspected area of the false certification, license plates achieve accurate positioning.
Experimental results show that the complexity of the proposed algorithm is low for many of the license plate image segmentation. It can meet the fast, accurate positioning requirements for LPD system. What’s more is that this algorithm has pretty good robustness due to this algorithm not very sensitive to most of the context of complex, uneven lighting, low contrast characters and background, as well as distribution vehicles.
引文
[1]孙兴征.车牌识别系统中的牌照定位分割技术研究[D].重庆:重庆大学,2004:
[2]沈世旻,盛翊智.基于边缘检测的车牌图像分割技术[J].自动化技术与应用,2004,23(4):24-26
[3]管庶安,廖明潮,万常芹.基于车牌字符笔画特征的快速定位算法[J].计算机工程与应用,2005,41(12):186-188.
[4]张于青,包劲松,金烨.一种基于多重特征的车牌定位算法[J].计算机工程与应用,2006,3(2):220-222.
[5]全书海,薛志华,王琴.基于投影图像分布特征的车牌定位算法研究与实现[J].武汉理工大学学报,2005,29(6):879-882
[6] T ALL A,J BULAS-CRUZ,E L DAGLESS.Vision Based Road Traffic Data Collection[A].Proc International Symposium on Automotive Technology and Automation(ISATA),26thInternational Conference[C].Aachen,Germany,1993:609-616
[7] JOHNSON A S,BIRD B M.Number-plate matching for automatic vehicle Identification.Electronic Image and Image Processing in Security and Forensic.IEEE Colloquium[C].1990(4):1-8
[8] R A LOTUFO,A D MORGAN,A S JOHNSON.Automatic unmber-plate recognition,Image Analysis for Transport Applications[J].IEE Colloquium,1990(6):1-6
[9] Young Sung Soh,Byung Tac Chun,HubYoon,Design of real time vehicle identification system,IEEE Intemational Confereneeon System,Man,and Cybemeties,Humans,Information and Technology 1994(3):2147-2152
[10]陈兆学,施鹏飞.基于灰度图像的车牌快速定位和分割方法.[J].计算机工程.2006,32(9):173-177
[11]韩永强,李世祥.基于线处理的子图像定位算法[J].计算机与现代化.1998(2):10-12
[12]魏武,黄心汉,王敏.一种基于垂直字符边界特征的车牌定位方法[J].中国公路学报.2000,13(4):88-99
[13]王庆,张炜,赵荣椿.基于自适应门限和宽线检测的车牌定位方法[J].中国体视学与图像分析.2000(2):65-70
[14]张海燕,王丽,张建军.一种基于小波的快速车牌定位算法研究[J].工业控制计算机.2006,19(1):63-64
[15]蒋先刚,吴小林,余鹤龄.基于灰色形态学的车牌定位技术研究[J].华东交通大学学报.2006,23(2):57-61
[16]张玲,刘勇,何伟.自适应遗传算法在车牌定位中的应用.计算机应用[J].2008,28(1):184-186
[17]马永力,夏秋华.基于数学形态学的车牌精定位算法的研究[J].微计算机信息.2008,24(1):227-229
[18]崔屹.图像处理与分析[M].北京:科学出版社,2000年:98.
[19]白洪亮,娄正良,邹明福,刘昌平.复杂背景下基于形态学的车牌识别系统[J].公路交通科技.2004,21(10):117-120
[20]卢雅琴,邬凌超.基于数学形态学的车牌定位方法[J].计算机工程.2005,31(3):224-226
[21]戴青云,余英林.一种基于小波与形态学的车牌图象分割方法[J].中国图形图象学报.2000,5A(5):411-415
[22]张向东,马月涛,杨奇泽.一种基于灰度图像的车牌快速定位算法[J].电子科技.2007,212(5):173-177
[23]王良红,王锦玲,梁延华.改进的Hough变换在校正汽车牌照倾斜中的应用[J].计算机工程.2004,2(1):45-48
[24]李皓迪,基于邻域灰度变化矢量场的一种图像边缘检测的新方法[D].成都:电子科技大学,2006:
[25]孙虹,方敏.一种彩色汽车车牌的自动分割方法[J].合肥工业大学学报.2005,28(11):1427-1430
[26] Park S.H.,Kim K.I.Jung K.,Kim H.J.,Locating car Iicense plalte susing networks, Electronic Letters,1999(35):1475-1477
[27]张引,潘云鹤.彩色汽车图像牌照定位新方法[J].中国图像图形学报.2001,6(4):374-377
[28]赵涛,杨晓莉,王绪本,张娜.一种用于车牌定位的改进BP神经网络方法[J].计算机仿真.2007,24(2):240-243
[29]段震.基于构造性学习的车牌定位与识别系统研究[D].安徽:安徽大学,2004:
[30]魏明,龚家伟,智能交通运输系统(ITS)及其发展现状[J].贵州大学学报(农业与生物科学版),2002,21(5):375-380
[31]穆长江.车辆牌照自动识别的算法研究[D].沈阳:沈阳工业大学,2004:
[32]宋晨光.车牌定位与字符分割算法研究[D].北京:中国农业大学,2005:
[33]左望霞.复杂背景下多车牌图像分割技术研究[D].成都:电子科技大学,2006:
[34]于殿泓.图像检测与处理技术[M].西安:西安电子科技大学出版社.2006年:125.
[35] P.Soille著.王小鹏等译.形态学图像分析原理与应用[M].北京:清华大学出版社.2008年:104.
[36]阮秋琦.数字图像处理学[M].北京:电子工业出版社.2001年:98.
[37]陈武凡.小波分析及其在图像处理中的应用[M].北京:科学出版社.2002年:201.
[38]杨辉,张继武.数学形态学在图像边缘检测中的应用研究[J].辽宁大学学报,2005,32(1):50-53
[39]罗军辉,冯平.MATLAB7.0在图像处理中的应用[M].北京:机械工业出版社.2006年:203.
[40] Jun-Wei Hsieh,Shih-Hao Yu, Yung-Sheng Chen.Morphology-based License Plate Detection from Complex Scenes[J].IEEE 2002:176-179
[41]包明,路小波.基于Hough变换的车牌倾斜检测算法[J].交通与计算机.2004,22(2):57-60
[42]李文举,梁德群,崔连延,王新年.一种新的车牌倾斜校正方法[J].信息与控制.2004,33(2):231-235
[43]王国良,梁德群.一种新的车牌定位与倾斜校正方法[J].计算机应用.2008,28(7):1890-1895