基于数字图像处理技术的车辆违章检测系统
|
摘要
迄今，汽车已经成为人们出行中最重要的交通工具，汽车的普及给人们生活带来便利，也带来了很多问题，如公路交通的压力、交通监管的难度以及人们出行的危险。基于此类问题，人们着手研发了基于数字图像处理技术的车辆违章检测系统。车辆违章检测系统是智能交通系统的重要组成部分，它的出现改善了日渐拥堵的道理交通。由于自然因素和人本身的生理因素，交通监管在时间和空间上存在一定的空白，如夜晚、恶劣的雨雪天气等。自动违章检测系统的出现，有效的填补了这个管理上的空当，解决了因为车辆和道路情况复杂带来的警力不足的问题。智能交通系统对中国未来和谐道路交通的发展有着深远的影响。
全球各地交通事故的研究数据显示，十字路口是各个国家交通事故发生的重灾区，而这其中超过半数的原因和司机违章闯红灯有关。这已经成为威胁道路交通和人们生命财产安全的重大隐患。传统的解决方法是在各交通路口安装感应线圈或车轴传感器。这种方法在一定时期内取得了不错的效果。但随着科技的发展，这种传统的方法本身的各种弊端和不足逐渐显现出来：对道路的破坏问题、施工上的不便以及施工时对交通产生的影响，频繁的检修、维护等，这种方式已经不再适合现今高密度、高效率的交通监管要求。采用基于数字图像处理的技术的自动交通检测系统，有效利用了计算机数字图像处理技术的成熟发展，较过去传统的方法，不论是在施工安装方面，还是在运行维护以及取得效果方面都有了全面的提高。采用图像处理技术，方便、快捷、高效，解决了传统设备老化带来的误差。
本文详细阐述了“基于数字图像处理的车辆违章检测系统”的结构、所采用的技术以及实现的功能。在深入研究了《中华人民共和国道路交通安全法》和《机动车违法自动记录系统通用技术条件》的基础上，系统的学习了3张照片的要求。根据法律规定，违反交通信号灯是指机动车在信号控制的交叉路口和路段上，红灯亮起时，车头越过停止线并继续行驶的行为。认定闯红灯，系统需记录机动车在3个不同位置的图像信息。第一副在越过停止线前拍摄，第二幅记录整个车身已经越过停止线，信号灯为红色的情况，第三幅是记录机动车明确违法行为后的行驶方向。
本文通过四个步骤来完成车辆违章闯红灯行为的识别，主要是目标图像信息的采集和重构、数字图像的预处理、运动目标的特征提取、违章行为的识别。图像预处理过程主要是灰度变换、直方图变换以及锐化和平滑去噪处理。这些处理的目的是去除图像在摄取和传递过程中产生的噪声和畸变，方便后续的观察和识别。之后对图像进行二值化处理，图像的增强、分割和边缘检测。本文的中心思想在于提取运动目标，再对目标的行为进行判断识别，我们只需要考虑我们需要的信息，通过二值化处理可以将数字图像分成前景和背景两部分，忽略掉我们不需要的一些杂质信息，如树木、栅栏等。在对运动目标的检测和提取过程中，使用基于三帧差分的运动目标检测算法，相对于传统的两帧差分法，三帧差分具有更好的抗噪声能力。三帧差分的优点是对相邻的两帧差图像再进行相与运算，这样可以使图像的轮廓更加清晰，突出前景目标，背景分割的效果也更好。在违章行为检测方面，主要采用了基于道路分割思想的k均值聚类算法，选择的5个相关属性分别是行进距离、行进角度、车型、质心X像素和质心Y像素。前三个属性分别从不同的角度描述车辆的行进状态，而后两个属性的目的是使目标对象的像素集中。
本文设计的宗旨在于找到更加实用、方便有效的车辆违章检测方法。课题是基于物联网和数字图像处理相关的设计，其中的通讯部分由本门师兄弟设计，所以在此只略作描述，论文的重点在于图像的处理方法和违章行为的识别。
Today, cars have become one of the most important vehicles while people travelling. Thepopularity of cars brings a lot of convenience to people's daily life, but also bringsdisadvantages at the same time, such as the heavy pressure of road traffic, the difficulty oftraffic supervision and the risk of travel. Due to these problems, people began to develop aVehicles Violation detection system based on digital image processing technology. Thissystem is an important part of the intelligent traffic detection systems, whose appearancegreatly improved the increasingly congested traffic conditions. There is a big blank areabased on the time and place conditions for traffic supervision, due to natural factors andhuman physiological factors, such as night hours, bad rainy weather, and snowy weather andso on. Therefore, the appearance of intelligent traffic detection system fills the supervisionblank efficiently, and solves the police shortage problem caused by the vehicle number andcomplexity of road condition. This system has a profound impact for the development ofChina's future harmonious road traffic.
According to the statistics of traffic accidents around the world, the crossroads becomethe hardest-hit area for all countries, and more than half of the reason for this is driversrunning red lights problem. This situation has become a significant threat against traffic safety,people's lives and property safety. The traditional solution is to fit induction coil sensors oraxle sensors at various traffic junctions. This method has achieved good results during acertain period of time. But With the development of science and technology, its variousdeficiencies and shortcomings emerge gradually: destruction problem to roads, constructioninconvenience and the impact for traffic while constructing, such as frequent checkout andmaintenance. This is no longer suitable for these high-density and high-efficiencymanagement requirements of today. Intelligent traffic detection systems based on digitalimage processing technology, uses the computer's advanced digital image tech effectively,have reached comprehensive improvement and not only in construction and installation butalso in maintenance and impact, compared to the traditional methods in past. Adopting imageprocessing tech, it solves the aging problem of the old equipment with convenience,speediness, and high efficiency.
This paper describes the structure of "Vehicles Violation detection system based ondigital image processing technology", the technology used and the functions achieved. AfterIn-depth study of the" People's Republic of China's Road Traffic Safety Law " and " motorvehicle violation automatic recording system's general technical conditions", learn the requirements of the three photos. Violation of traffic lights means that motor vehicle crossesthe stop-line and continue driving when the red-light at a signal-controlled intersections andsections. Red light running identified, the system needs to record motor vehicle images inthree different locations. First one is taken before crossing stop line, the second is taken whenthe whole body of the vehicle has crossed the stop line while light red, the third one is takenwhile motor vehicles entering the intersection or cornering to ensure the driving direction.
This paper uses four steps to complete the process of identifying vehicles running redlights, they are: acquisition and reconstruction of target image, pre-processing of digitalimages, extracting characters of moving objects, and the identification of violations. Theimage preprocessing process is mainly about gray level transformation; histogramtransformation as well as sharpening and smoothing denoise. The purpose is to remove imagenoise and distortion in the uptake and delivery process, to facilitate follow-up observation andidentification. After that are image binarization processing, image enhancement, segmentationand edge detection. The central idea of this paper is to extract moving target, and then tojudge the behavior of the target, we could only consider the information we need. Thedigital image is divided into two parts of foreground and background by binarizationprocessing, ignoring the mussy information we do not need, such as trees, fences, etc. In theprocess of detection and extraction moving target, use the moving target detection algorithmbased on three-temporal differencing, which has better noise immunity than traditionaltwo-temporal differencing. The advantage of the three-temporal differencing is doing logicalAND(&&) computing between two adjacent temporal, this could make the outline of theimage more clear, project foreground objects and background segmentation results would bebetter. As for violations detection, it mainly uses K-average clustering algorithm based onroad segmentation thinking, the selected five attributes are the distance traveled, the angletraveled, car models, the centroid of X pixels and centroid of Y pixels. The first threeproperties are used to describe the moving state of the vehicle from different angles; thepurpose of the last two properties is to concentrate the pixel of the target object.
In this paper, the design aims to find a more practical, convenient and effective vehicleviolation detection methods. The subject is designed based on the Internet of Things andrelated to the digital image processing, in which the communication part is implemented bymy senior fellow apprentice, so only a brief description about this part. This paper focuses onimage processing methods and the identification of the violation behavior.
全球各地交通事故的研究数据显示，十字路口是各个国家交通事故发生的重灾区，而这其中超过半数的原因和司机违章闯红灯有关。这已经成为威胁道路交通和人们生命财产安全的重大隐患。传统的解决方法是在各交通路口安装感应线圈或车轴传感器。这种方法在一定时期内取得了不错的效果。但随着科技的发展，这种传统的方法本身的各种弊端和不足逐渐显现出来：对道路的破坏问题、施工上的不便以及施工时对交通产生的影响，频繁的检修、维护等，这种方式已经不再适合现今高密度、高效率的交通监管要求。采用基于数字图像处理的技术的自动交通检测系统，有效利用了计算机数字图像处理技术的成熟发展，较过去传统的方法，不论是在施工安装方面，还是在运行维护以及取得效果方面都有了全面的提高。采用图像处理技术，方便、快捷、高效，解决了传统设备老化带来的误差。
本文详细阐述了“基于数字图像处理的车辆违章检测系统”的结构、所采用的技术以及实现的功能。在深入研究了《中华人民共和国道路交通安全法》和《机动车违法自动记录系统通用技术条件》的基础上，系统的学习了3张照片的要求。根据法律规定，违反交通信号灯是指机动车在信号控制的交叉路口和路段上，红灯亮起时，车头越过停止线并继续行驶的行为。认定闯红灯，系统需记录机动车在3个不同位置的图像信息。第一副在越过停止线前拍摄，第二幅记录整个车身已经越过停止线，信号灯为红色的情况，第三幅是记录机动车明确违法行为后的行驶方向。
本文通过四个步骤来完成车辆违章闯红灯行为的识别，主要是目标图像信息的采集和重构、数字图像的预处理、运动目标的特征提取、违章行为的识别。图像预处理过程主要是灰度变换、直方图变换以及锐化和平滑去噪处理。这些处理的目的是去除图像在摄取和传递过程中产生的噪声和畸变，方便后续的观察和识别。之后对图像进行二值化处理，图像的增强、分割和边缘检测。本文的中心思想在于提取运动目标，再对目标的行为进行判断识别，我们只需要考虑我们需要的信息，通过二值化处理可以将数字图像分成前景和背景两部分，忽略掉我们不需要的一些杂质信息，如树木、栅栏等。在对运动目标的检测和提取过程中，使用基于三帧差分的运动目标检测算法，相对于传统的两帧差分法，三帧差分具有更好的抗噪声能力。三帧差分的优点是对相邻的两帧差图像再进行相与运算，这样可以使图像的轮廓更加清晰，突出前景目标，背景分割的效果也更好。在违章行为检测方面，主要采用了基于道路分割思想的k均值聚类算法，选择的5个相关属性分别是行进距离、行进角度、车型、质心X像素和质心Y像素。前三个属性分别从不同的角度描述车辆的行进状态，而后两个属性的目的是使目标对象的像素集中。
本文设计的宗旨在于找到更加实用、方便有效的车辆违章检测方法。课题是基于物联网和数字图像处理相关的设计，其中的通讯部分由本门师兄弟设计，所以在此只略作描述，论文的重点在于图像的处理方法和违章行为的识别。
Today, cars have become one of the most important vehicles while people travelling. Thepopularity of cars brings a lot of convenience to people's daily life, but also bringsdisadvantages at the same time, such as the heavy pressure of road traffic, the difficulty oftraffic supervision and the risk of travel. Due to these problems, people began to develop aVehicles Violation detection system based on digital image processing technology. Thissystem is an important part of the intelligent traffic detection systems, whose appearancegreatly improved the increasingly congested traffic conditions. There is a big blank areabased on the time and place conditions for traffic supervision, due to natural factors andhuman physiological factors, such as night hours, bad rainy weather, and snowy weather andso on. Therefore, the appearance of intelligent traffic detection system fills the supervisionblank efficiently, and solves the police shortage problem caused by the vehicle number andcomplexity of road condition. This system has a profound impact for the development ofChina's future harmonious road traffic.
According to the statistics of traffic accidents around the world, the crossroads becomethe hardest-hit area for all countries, and more than half of the reason for this is driversrunning red lights problem. This situation has become a significant threat against traffic safety,people's lives and property safety. The traditional solution is to fit induction coil sensors oraxle sensors at various traffic junctions. This method has achieved good results during acertain period of time. But With the development of science and technology, its variousdeficiencies and shortcomings emerge gradually: destruction problem to roads, constructioninconvenience and the impact for traffic while constructing, such as frequent checkout andmaintenance. This is no longer suitable for these high-density and high-efficiencymanagement requirements of today. Intelligent traffic detection systems based on digitalimage processing technology, uses the computer's advanced digital image tech effectively,have reached comprehensive improvement and not only in construction and installation butalso in maintenance and impact, compared to the traditional methods in past. Adopting imageprocessing tech, it solves the aging problem of the old equipment with convenience,speediness, and high efficiency.
This paper describes the structure of "Vehicles Violation detection system based ondigital image processing technology", the technology used and the functions achieved. AfterIn-depth study of the" People's Republic of China's Road Traffic Safety Law " and " motorvehicle violation automatic recording system's general technical conditions", learn the requirements of the three photos. Violation of traffic lights means that motor vehicle crossesthe stop-line and continue driving when the red-light at a signal-controlled intersections andsections. Red light running identified, the system needs to record motor vehicle images inthree different locations. First one is taken before crossing stop line, the second is taken whenthe whole body of the vehicle has crossed the stop line while light red, the third one is takenwhile motor vehicles entering the intersection or cornering to ensure the driving direction.
This paper uses four steps to complete the process of identifying vehicles running redlights, they are: acquisition and reconstruction of target image, pre-processing of digitalimages, extracting characters of moving objects, and the identification of violations. Theimage preprocessing process is mainly about gray level transformation; histogramtransformation as well as sharpening and smoothing denoise. The purpose is to remove imagenoise and distortion in the uptake and delivery process, to facilitate follow-up observation andidentification. After that are image binarization processing, image enhancement, segmentationand edge detection. The central idea of this paper is to extract moving target, and then tojudge the behavior of the target, we could only consider the information we need. Thedigital image is divided into two parts of foreground and background by binarizationprocessing, ignoring the mussy information we do not need, such as trees, fences, etc. In theprocess of detection and extraction moving target, use the moving target detection algorithmbased on three-temporal differencing, which has better noise immunity than traditionaltwo-temporal differencing. The advantage of the three-temporal differencing is doing logicalAND(&&) computing between two adjacent temporal, this could make the outline of theimage more clear, project foreground objects and background segmentation results would bebetter. As for violations detection, it mainly uses K-average clustering algorithm based onroad segmentation thinking, the selected five attributes are the distance traveled, the angletraveled, car models, the centroid of X pixels and centroid of Y pixels. The first threeproperties are used to describe the moving state of the vehicle from different angles; thepurpose of the last two properties is to concentrate the pixel of the target object.
In this paper, the design aims to find a more practical, convenient and effective vehicleviolation detection methods. The subject is designed based on the Internet of Things andrelated to the digital image processing, in which the communication part is implemented bymy senior fellow apprentice, so only a brief description about this part. This paper focuses onimage processing methods and the identification of the violation behavior.
引文
[1]王耀南,李树涛,毛建旭.计算机图像处理与识别技术[M].北京:高等教育出版社，2001.
[2]于万波.基于MATLAB的图像处理[M].北京:清华大学出版社,2008.
[3]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.
[4]陆润民.计算机图形学教程[M].北京:清华大学出版社,2003.
[5]韩培友,董桂云.图像技术[M].西安:西北工业大学出版社,2009.
[6]杨高波,杜青松.图像视频处理应用及实例[M].北京:电子工业出版社,2010.
[7]夏德深,傅德胜.计算机图像处理及应用实验教程[M].南京:东南大学出版社,2005.
[8]刘丹.计算机图像处理的数学和算法基础[M].北京:国防工业出版社,2005.
[9]孙即祥.图像处理[M].北京:科学出版社,2009.
[10]闫敬文.数字图像处理技术[M].北京:国防工业出版社,2007.
[11]唐波,马伯宁,鲁敏.计算机图形图像处理基础[M].北京:电子工业出版社,2011.
[12]（美）普拉特著,张引等译.数字图像处理[M].北京:机械工业出版社,2009.
[13]吕凤军.数字图像处理编程入门[M].北京:清华大学出版社,2000.
[14]周长发.精通Visual C++图像编程[M].北京:电子工业出版社,2000.
[15]伍俊良. VC++课程设计与系统开发[M].北京:清华大学出版社,2002.
[16]杨淑莹. VC++图像处理程序设计[M].北京:清华大学出版社,2009.
[17]田春秀行著,乔双译.计算机图像处理[M].北京:科学出版社,2004.
[18]崔屹.数字图像处理技术与应用[M].北京:电子工业出版社,1997.
[19]陈兵旗,孙明. Visual C++实用图像处理[M].北京:清华大学出版社,2004.
[20]林锐.高质量C++/C编程指南[M].北京:电子工业出版社,2002
[21] JianFeng Xu, Shaofa Li, Mianshui Yu. Car License Plate Extraction Using Color andEdge information[C]. Proceedings of the Third International Conference on MachineLearning and Cybernetics. Shanghai.2004:26-29.
[22] R.Parisi, et al. Car Plate Recognition by Neual and Image Processing[C]. IEEE ISCAS.USA.1998:195-198.
[23] Charl Coetzee, et al. PC Based Number Plate Recognition Systems[C]. In Proc IEEEInternational Symposium on Industrial Electronics.1998:605-610.
[24] Weigang Zhu, Guojiang Hou, Xing Jia. A Study of Locating Vehicle Plate Based onColor Feature and Mathematical Morphology[J]. Signal Processing.2002,1:748:751.
[25] Alpert S, Galun M, Basri R, Brandt A. Image segmentation by probabilistic bottom-upaggregation and cue integration[C]. In: CVPR (2007).
[26] Zelnik-manor L, Perona P. Self-tuning spectral clustering[C].In: NIPS(2004),1601-1608.
[27] Giovanni Adomi, Stefano Cagni, Macro Gori, Monica Mordonini. Access ControlSystem with Neuro Fuzzy Supervision[C]. IEEE Intelligent Transportation SystemsConference Proceedings.2001:472-477.
[28] Pushkin Kachroo. Real-time Travel Time Estimation using Macroscopic Traffic FlowModels[C].2001IEEE Intelligent Transportation Systems Conference ProceedingOakland (CA), USA-August25-29,2001:132-137.
[29]郭悦.基于三帧差分和Cam-shift改进法的运动目标检测与跟踪研究[D].燕山大学理学院,2011.
[30] D.M.Emiris, D.E.Koulouriotis. Automated Optic Recognition of Alphanumeric Contentin Car License Plates in a Semi-structured Environment[J]. IEEE Transaction on PatternAnalysis and Machine Intelligence.2001,7(8):121-129.
[31]王小辉.闯红灯事件的视频检测算法的研究和应用[D].厦门:厦门大学计算机应用技术学院,2008.
[32] Nenad Stanic, et al. Character Recognition Using a Cellular Neural Network[J]. IEEENeural.2002:135-138.
[33] Junwei Hsieh, Shihao Yu, Yunsheng Chen. Morphology-Based License Plate Detectionfrom Complex Scenes[J]. Pattern Recognition.2002:176-179.
[34] R. Mech, M. Wollbom. A Noise Robust Method for Segmentation of Moving Objects inVideo Sequences[C]. Proc IEEE International Conference on Acoustics, Speech andSignal Processing,2006:41-45.
[35] A. J. Lipton, H. Fujiyoshi, R. S. Patil. Moving Target Classification and Tracking fromReal-time Video[C]. Proc IEEE Workshop on Applications of Computer Vision,2007:8-14.
[2]于万波.基于MATLAB的图像处理[M].北京:清华大学出版社,2008.
[3]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.
[4]陆润民.计算机图形学教程[M].北京:清华大学出版社,2003.
[5]韩培友,董桂云.图像技术[M].西安:西北工业大学出版社,2009.
[6]杨高波,杜青松.图像视频处理应用及实例[M].北京:电子工业出版社,2010.
[7]夏德深,傅德胜.计算机图像处理及应用实验教程[M].南京:东南大学出版社,2005.
[8]刘丹.计算机图像处理的数学和算法基础[M].北京:国防工业出版社,2005.
[9]孙即祥.图像处理[M].北京:科学出版社,2009.
[10]闫敬文.数字图像处理技术[M].北京:国防工业出版社,2007.
[11]唐波,马伯宁,鲁敏.计算机图形图像处理基础[M].北京:电子工业出版社,2011.
[12]（美）普拉特著,张引等译.数字图像处理[M].北京:机械工业出版社,2009.
[13]吕凤军.数字图像处理编程入门[M].北京:清华大学出版社,2000.
[14]周长发.精通Visual C++图像编程[M].北京:电子工业出版社,2000.
[15]伍俊良. VC++课程设计与系统开发[M].北京:清华大学出版社,2002.
[16]杨淑莹. VC++图像处理程序设计[M].北京:清华大学出版社,2009.
[17]田春秀行著,乔双译.计算机图像处理[M].北京:科学出版社,2004.
[18]崔屹.数字图像处理技术与应用[M].北京:电子工业出版社,1997.
[19]陈兵旗,孙明. Visual C++实用图像处理[M].北京:清华大学出版社,2004.
[20]林锐.高质量C++/C编程指南[M].北京:电子工业出版社,2002
[21] JianFeng Xu, Shaofa Li, Mianshui Yu. Car License Plate Extraction Using Color andEdge information[C]. Proceedings of the Third International Conference on MachineLearning and Cybernetics. Shanghai.2004:26-29.
[22] R.Parisi, et al. Car Plate Recognition by Neual and Image Processing[C]. IEEE ISCAS.USA.1998:195-198.
[23] Charl Coetzee, et al. PC Based Number Plate Recognition Systems[C]. In Proc IEEEInternational Symposium on Industrial Electronics.1998:605-610.
[24] Weigang Zhu, Guojiang Hou, Xing Jia. A Study of Locating Vehicle Plate Based onColor Feature and Mathematical Morphology[J]. Signal Processing.2002,1:748:751.
[25] Alpert S, Galun M, Basri R, Brandt A. Image segmentation by probabilistic bottom-upaggregation and cue integration[C]. In: CVPR (2007).
[26] Zelnik-manor L, Perona P. Self-tuning spectral clustering[C].In: NIPS(2004),1601-1608.
[27] Giovanni Adomi, Stefano Cagni, Macro Gori, Monica Mordonini. Access ControlSystem with Neuro Fuzzy Supervision[C]. IEEE Intelligent Transportation SystemsConference Proceedings.2001:472-477.
[28] Pushkin Kachroo. Real-time Travel Time Estimation using Macroscopic Traffic FlowModels[C].2001IEEE Intelligent Transportation Systems Conference ProceedingOakland (CA), USA-August25-29,2001:132-137.
[29]郭悦.基于三帧差分和Cam-shift改进法的运动目标检测与跟踪研究[D].燕山大学理学院,2011.
[30] D.M.Emiris, D.E.Koulouriotis. Automated Optic Recognition of Alphanumeric Contentin Car License Plates in a Semi-structured Environment[J]. IEEE Transaction on PatternAnalysis and Machine Intelligence.2001,7(8):121-129.
[31]王小辉.闯红灯事件的视频检测算法的研究和应用[D].厦门:厦门大学计算机应用技术学院,2008.
[32] Nenad Stanic, et al. Character Recognition Using a Cellular Neural Network[J]. IEEENeural.2002:135-138.
[33] Junwei Hsieh, Shihao Yu, Yunsheng Chen. Morphology-Based License Plate Detectionfrom Complex Scenes[J]. Pattern Recognition.2002:176-179.
[34] R. Mech, M. Wollbom. A Noise Robust Method for Segmentation of Moving Objects inVideo Sequences[C]. Proc IEEE International Conference on Acoustics, Speech andSignal Processing,2006:41-45.
[35] A. J. Lipton, H. Fujiyoshi, R. S. Patil. Moving Target Classification and Tracking fromReal-time Video[C]. Proc IEEE Workshop on Applications of Computer Vision,2007:8-14.