基于复杂背景噪声的汽车车型识别研究
|
摘要
在科技高速发展的今天,计算机技术、电子、信息、通信以及自动化被普遍应用到各个领域,其中也包括道路交通管理。为了解决道路交通管理、路桥收费站自动收费等问题,世界各国纷纷把目光投向了对各种交通运输系统的研究开发。这些交通系统的应用可解除收费站的“瓶颈”制约作用,较好地缓解收费站的交通拥挤、排队现象,侦察违章车辆、处理交通事故等。目前已有一些比较成熟的系统,并且很多已经投入使用,但是这些系统存在着一些局限性,也有的系统是在半人工半自动化的情况下操作的。本文的选题就是在这样一个大的研究背景下提出的。
本文从汽车收费站实地拍摄的汽车图片着手研究,发现用于车型识别的汽车图片通常带有极复杂的背景噪声,而这些背景噪声不可能通过诸如背景相减等一些简单的方法去除掉。在本文研究过程的前期,遵循了提取轮廓线、寻找识别特征信息以识别车型的步骤。在提取轮廓线的方法研究中,首先把多种较成熟的方法应用到实验中,比较各种方法在本研究课题中的优劣,最后再把地理中等高线的原理应用到图像处理中,得到了较理想的车型轮廓,解决了在单一背景下提取轮廓线的局限性问题。用直线拟合车型轮廓曲线,不同的车型会有不同直线斜率和成角信息,由此提出了斜率之差算法,用于车型轮廓等高线的直线拟合。该算法为后续的车型分类识别提供了非常有用和可靠的特征样本信息。
提取的车型轮廓是车型识别的一个重要特征,在此基础上,本文也关注到了汽车的长、宽、高等基本特征信息,特别是两车胎中心距这个特征成为本论文中车型识别的重要参数。通过多种方法的实验和比较,最后,文中抛弃了原有去除复杂背景噪声的想法,从汽车对象本身出发:首先,把等高线的原理应用到汽车模板的定位中;用Canny边缘检测器检测出目标车图和标准车图的边缘信息,并对标准车图的边缘图像用贴标签法得到各种车型的模板;最后,用模板匹配的算法,把各种车型的模板和目标车图做匹配,以达到识别各种车型的目的。实验证明,此方法可得到较高的识别精度。
With the rapid development of science and technology, Computer technology, electronics, information, communications and automation is widely applied to various fields, in which also include the road traffic management. In order to solve the problems in the road traffic management, Countries around the world are eyeing the various transport systems in research and development. Application of these transportation systems able to remove the Bottleneck of toll station, better ease traffic congestion and queuing phenomenon of the toll station, reconnaissance illegal vehicles and deal with traffic accidents. At present, some mature systems had been used, but these are some limitations in these systems, and some systems are semi-automatic. In brief, in such a large research background, the work which we need to do are also very much.
The research is to recognize vehicles from photographs shooting in the toll station. There are complex background noises in the photographs used to recognize vehicles. But, these complex background noises can’t be wiped off used some simple methods such as the background cancellation. So, the goal to recognize vehicle can’t achieve. In the earlier period of the research, we followed the steps of extraction contour, searching character of the contour to achieve vehicle recognition. Many kinds of method had been used in the research of the extraction contour. We compared with each method in this research. Finally, contour principle in geography was applied in the image processing to get the contours of vehicle. It is shown that: the principle of Contour used to get the contours of vehicle would gain more perfectly effect. The limitation to get the contours of vehicle in the single background was solved by author. Based on different slope and angle information on different contours of vehicles, the arithmetic was brought forward into fitting beeline with difference of slope, which would provide usable and credible information to the later distinguishing of the vehicle contours.
The contour of vehicle is an important characteristic of vehicle recognition. The length and width of vehicle also is paid attention, and the center of two tires’and the distance of the two centers became an important parameter of the vehicle recognition. Through experiment and comparison of many kinds of method, the idea to wipe off complex background noise was given up. Firstly, the principle of Contour was used to position the templates of vehicles in this article. The edge detector of Canny operator was used to detect the edge of object vehicle and standard vehicle, and the method of label was used to get the templates of vehicles. Finally, the templates were used to match the object vehicle based on the arithmetic of template matching. It is shown that: this method would attain the intention of recognition of vehicles, and would gain more perfectly precision.
本文从汽车收费站实地拍摄的汽车图片着手研究,发现用于车型识别的汽车图片通常带有极复杂的背景噪声,而这些背景噪声不可能通过诸如背景相减等一些简单的方法去除掉。在本文研究过程的前期,遵循了提取轮廓线、寻找识别特征信息以识别车型的步骤。在提取轮廓线的方法研究中,首先把多种较成熟的方法应用到实验中,比较各种方法在本研究课题中的优劣,最后再把地理中等高线的原理应用到图像处理中,得到了较理想的车型轮廓,解决了在单一背景下提取轮廓线的局限性问题。用直线拟合车型轮廓曲线,不同的车型会有不同直线斜率和成角信息,由此提出了斜率之差算法,用于车型轮廓等高线的直线拟合。该算法为后续的车型分类识别提供了非常有用和可靠的特征样本信息。
提取的车型轮廓是车型识别的一个重要特征,在此基础上,本文也关注到了汽车的长、宽、高等基本特征信息,特别是两车胎中心距这个特征成为本论文中车型识别的重要参数。通过多种方法的实验和比较,最后,文中抛弃了原有去除复杂背景噪声的想法,从汽车对象本身出发:首先,把等高线的原理应用到汽车模板的定位中;用Canny边缘检测器检测出目标车图和标准车图的边缘信息,并对标准车图的边缘图像用贴标签法得到各种车型的模板;最后,用模板匹配的算法,把各种车型的模板和目标车图做匹配,以达到识别各种车型的目的。实验证明,此方法可得到较高的识别精度。
With the rapid development of science and technology, Computer technology, electronics, information, communications and automation is widely applied to various fields, in which also include the road traffic management. In order to solve the problems in the road traffic management, Countries around the world are eyeing the various transport systems in research and development. Application of these transportation systems able to remove the Bottleneck of toll station, better ease traffic congestion and queuing phenomenon of the toll station, reconnaissance illegal vehicles and deal with traffic accidents. At present, some mature systems had been used, but these are some limitations in these systems, and some systems are semi-automatic. In brief, in such a large research background, the work which we need to do are also very much.
The research is to recognize vehicles from photographs shooting in the toll station. There are complex background noises in the photographs used to recognize vehicles. But, these complex background noises can’t be wiped off used some simple methods such as the background cancellation. So, the goal to recognize vehicle can’t achieve. In the earlier period of the research, we followed the steps of extraction contour, searching character of the contour to achieve vehicle recognition. Many kinds of method had been used in the research of the extraction contour. We compared with each method in this research. Finally, contour principle in geography was applied in the image processing to get the contours of vehicle. It is shown that: the principle of Contour used to get the contours of vehicle would gain more perfectly effect. The limitation to get the contours of vehicle in the single background was solved by author. Based on different slope and angle information on different contours of vehicles, the arithmetic was brought forward into fitting beeline with difference of slope, which would provide usable and credible information to the later distinguishing of the vehicle contours.
The contour of vehicle is an important characteristic of vehicle recognition. The length and width of vehicle also is paid attention, and the center of two tires’and the distance of the two centers became an important parameter of the vehicle recognition. Through experiment and comparison of many kinds of method, the idea to wipe off complex background noise was given up. Firstly, the principle of Contour was used to position the templates of vehicles in this article. The edge detector of Canny operator was used to detect the edge of object vehicle and standard vehicle, and the method of label was used to get the templates of vehicles. Finally, the templates were used to match the object vehicle based on the arithmetic of template matching. It is shown that: this method would attain the intention of recognition of vehicles, and would gain more perfectly precision.
引文
[1]王艳春,李建军,何鹏,尹明.公路交通管理中行驶车辆自动识别技术研究[J].微计算机信息(嵌入式与SOC),2006.22(1-2).
[2]徐建闽,贺敬凯.车型与车牌自动识别技术分析[J].交通与计算机, 2002(2).
[3]吴志攀,赵跃龙.车型图像分割方法的研究[J].惠州学院学报(自然科学版), 2005(12).
[4] (日)谷口庆治编.数字图像处理[M].北京:科学出版社, 2002.7.
[5] (美) Rafael C. Gonzalez, Richard E. Woods, Steven L. Eddins著阮秋琦等译.数字图像处理[M].北京:电子工业出版社, 2005.
[6]胡小锋,赵辉等编著.Visual C++/MATLAB图像处理与识别实用案例精选[M].北京:人民邮电出版社, 2004.9.
[7]刘学质.等高线方法及其应用[J].数学通讯.,2005(3).
[8]宋焕生,赵祥模,王养利.一种新的汽车牌照识别的图像增强算法[J].长安大学学报(自然科学版), 2006.26(4).
[9]宋丽,李令奇.ETC中具有车型识别功能的车辆检测器设计[J].微计算机信息(嵌入式与SOC), 2005, 21(8-2).
[10]贾新宇,陆志一.Polyline线的抽稀与圆滑[J].信息技术, 2006(7) .
[11]朱志勇,刘伟铭,伍友龙.基于Bayes法则和BP神经网络的高速动态情形下车型识别[J].计算机测量与控制, 2005.13(7).
[12]刘怡光,游志胜.一种用于图像目标识别的神经网络及其车型识别应用[J].计算机工程,2003(3).
[13]宋丹,徐蔚鸿.基于模糊理论的车型识别[J].计算机技术与发展, 2006.16(3).
[14]刘直芳,游志胜等.基于视觉模型的车型动态定位[J].计算机工程与应用, 2003.8.
[15]陈爱斌.基于特征车的汽车车型识别[J].信息技术, 2004.28(5).
[16]王亮申,朱玉才,陈少华等.利用SVM进行车型识别[J].计算机工程与设计, 2005.26(9).
[17] LI Wen-ju,LIANG De-qun,ZHANG Qi,et al.A novel approach for vehicle license plate location based on edge color pair[J]. Chinese Journal of Computers, 2004,27(2):204-208.
[18] CHANG Shyang-lih,CHEN Li-shien,CHUNG Yun-chung, et al. Automatic license plate recognition[J]. IEEE Tran on ITS, 2004,5(1):42-53.
[19] Chen Xi-lin,Zhang Jing,Alex W. Automatic detection and recognition of signs from nature scenes[J]. IEEE Trans on Image Processing, 2004,13(1):87-99.
[20]赵英男,刘正东,杨静宇.一种基于Gabor滤波器的车型识别方法[J].计算机工程, 2005.31(22).
[21]林亚明.一种基于EMD非相似距离度量的车型自动识别新方法[J].福建电脑,2005(9).
[22] (美)迪达(Duda,R.O.)等著李宏东等译.模式分类[M].北京:机械工业出版社, 2003.9.
[23]罗军辉,冯平,哈力旦·A等编著.MATLAB 7.0在图像处理中的应用[M].北京:机械工业出版社, 2005.
[24] Song H. The high performance car license plate recognition system and its core techniques[C] // ICVES: Pro-ceedings of IEEE int con on ICVES. Xi.an: IEEE ICVES.05, 2005:42-45.
[25]左奇,史忠科.一种基于数学形态学的实时车牌图象分割方法[J].中国图象图形学报(A版), 2003.8(3).
[26]赵海燕.利用改进的边缘算子进行车牌定位[J].长春理工大学学报, 2005.28(1).
[27] Jiangping Fan, Walid G Aref. An improved automatic isotropic color edge detection technique [J]. Pattern Recognition Letters, 2001, 22(6): 1419- 1429.
[28] Cheng H D, Jiang X H. Color image segmentation: Advances and prospects [J]. Pattern Recognition, 2001, 34(9): 2259-2281.
[29]林培群,徐建闽,傅惠,梁俊斌.多分支BP网络模型及其在车型分类中的应用[J].微计算机信息(嵌入式与SOC)[J], 2005,21(12-2).
[30]运国莲,陈启美.基于非完备贝叶斯网络的车型识别方法[J].交通与计算机, 2006, 24(1).
[31] Li Pei-hong, Ding Ya-li, Liu Ji-Lin. A video-based traffic information extraction system[J]. Proc IEEE:Intel Veh SymP, 2003, 4(l):523-528.
[32] Phillips P J, Moon H, Syed A, et al. The FERET evaluation methodology for face-recognition algorithms[J]. IEEE TPAMI, 2000, 22(10):1090-1104.
[33] Lienhart R, Liang L, Kuranov A. A detector tree of boosted classifiers for real-time object detection and tracking[J]. IEEE ICME, 2003,2(7):277-280.
[34] Zitova B, Flusser J, Image registration methods: a survey[J]. Image Vis. Comput. 2003,21:977-1000.
[35] Murat Balci, Hassan Foroosh. Subpixel estimation of shifts directly in the fourier domain[J]. IEEE Transaction on Image Processiong, 2006,15(7):123-124.
[36] Zhang G L, Ren X Y, Zhang T X, et al. Correlation Tracking Using a Novel Distance Measurement as Feedback[J]. Infrared and Laser Engineering,2003,32(6):624-629.
[2]徐建闽,贺敬凯.车型与车牌自动识别技术分析[J].交通与计算机, 2002(2).
[3]吴志攀,赵跃龙.车型图像分割方法的研究[J].惠州学院学报(自然科学版), 2005(12).
[4] (日)谷口庆治编.数字图像处理[M].北京:科学出版社, 2002.7.
[5] (美) Rafael C. Gonzalez, Richard E. Woods, Steven L. Eddins著阮秋琦等译.数字图像处理[M].北京:电子工业出版社, 2005.
[6]胡小锋,赵辉等编著.Visual C++/MATLAB图像处理与识别实用案例精选[M].北京:人民邮电出版社, 2004.9.
[7]刘学质.等高线方法及其应用[J].数学通讯.,2005(3).
[8]宋焕生,赵祥模,王养利.一种新的汽车牌照识别的图像增强算法[J].长安大学学报(自然科学版), 2006.26(4).
[9]宋丽,李令奇.ETC中具有车型识别功能的车辆检测器设计[J].微计算机信息(嵌入式与SOC), 2005, 21(8-2).
[10]贾新宇,陆志一.Polyline线的抽稀与圆滑[J].信息技术, 2006(7) .
[11]朱志勇,刘伟铭,伍友龙.基于Bayes法则和BP神经网络的高速动态情形下车型识别[J].计算机测量与控制, 2005.13(7).
[12]刘怡光,游志胜.一种用于图像目标识别的神经网络及其车型识别应用[J].计算机工程,2003(3).
[13]宋丹,徐蔚鸿.基于模糊理论的车型识别[J].计算机技术与发展, 2006.16(3).
[14]刘直芳,游志胜等.基于视觉模型的车型动态定位[J].计算机工程与应用, 2003.8.
[15]陈爱斌.基于特征车的汽车车型识别[J].信息技术, 2004.28(5).
[16]王亮申,朱玉才,陈少华等.利用SVM进行车型识别[J].计算机工程与设计, 2005.26(9).
[17] LI Wen-ju,LIANG De-qun,ZHANG Qi,et al.A novel approach for vehicle license plate location based on edge color pair[J]. Chinese Journal of Computers, 2004,27(2):204-208.
[18] CHANG Shyang-lih,CHEN Li-shien,CHUNG Yun-chung, et al. Automatic license plate recognition[J]. IEEE Tran on ITS, 2004,5(1):42-53.
[19] Chen Xi-lin,Zhang Jing,Alex W. Automatic detection and recognition of signs from nature scenes[J]. IEEE Trans on Image Processing, 2004,13(1):87-99.
[20]赵英男,刘正东,杨静宇.一种基于Gabor滤波器的车型识别方法[J].计算机工程, 2005.31(22).
[21]林亚明.一种基于EMD非相似距离度量的车型自动识别新方法[J].福建电脑,2005(9).
[22] (美)迪达(Duda,R.O.)等著李宏东等译.模式分类[M].北京:机械工业出版社, 2003.9.
[23]罗军辉,冯平,哈力旦·A等编著.MATLAB 7.0在图像处理中的应用[M].北京:机械工业出版社, 2005.
[24] Song H. The high performance car license plate recognition system and its core techniques[C] // ICVES: Pro-ceedings of IEEE int con on ICVES. Xi.an: IEEE ICVES.05, 2005:42-45.
[25]左奇,史忠科.一种基于数学形态学的实时车牌图象分割方法[J].中国图象图形学报(A版), 2003.8(3).
[26]赵海燕.利用改进的边缘算子进行车牌定位[J].长春理工大学学报, 2005.28(1).
[27] Jiangping Fan, Walid G Aref. An improved automatic isotropic color edge detection technique [J]. Pattern Recognition Letters, 2001, 22(6): 1419- 1429.
[28] Cheng H D, Jiang X H. Color image segmentation: Advances and prospects [J]. Pattern Recognition, 2001, 34(9): 2259-2281.
[29]林培群,徐建闽,傅惠,梁俊斌.多分支BP网络模型及其在车型分类中的应用[J].微计算机信息(嵌入式与SOC)[J], 2005,21(12-2).
[30]运国莲,陈启美.基于非完备贝叶斯网络的车型识别方法[J].交通与计算机, 2006, 24(1).
[31] Li Pei-hong, Ding Ya-li, Liu Ji-Lin. A video-based traffic information extraction system[J]. Proc IEEE:Intel Veh SymP, 2003, 4(l):523-528.
[32] Phillips P J, Moon H, Syed A, et al. The FERET evaluation methodology for face-recognition algorithms[J]. IEEE TPAMI, 2000, 22(10):1090-1104.
[33] Lienhart R, Liang L, Kuranov A. A detector tree of boosted classifiers for real-time object detection and tracking[J]. IEEE ICME, 2003,2(7):277-280.
[34] Zitova B, Flusser J, Image registration methods: a survey[J]. Image Vis. Comput. 2003,21:977-1000.
[35] Murat Balci, Hassan Foroosh. Subpixel estimation of shifts directly in the fourier domain[J]. IEEE Transaction on Image Processiong, 2006,15(7):123-124.
[36] Zhang G L, Ren X Y, Zhang T X, et al. Correlation Tracking Using a Novel Distance Measurement as Feedback[J]. Infrared and Laser Engineering,2003,32(6):624-629.