基于视频的车辆测速系统研究
|
摘要
Intelligent Transport System简称ITS,也就是智能交通系统。它是一种包含信息技术、数据通讯技术、电子传感技术、电子控制技术以及计算机处理技术等多技术的系统。它是一种能够在大范围内、全方位发挥作用的,实时、较准确、较高效的综合运输和管理系统。目前我国的智能交通系统已经在一些大城市得到了推广应用,尚未得到全面的普及。图像处理技术的迅速发展,势必将对ITS的发展做出一个巨大的推动。本文就是在这样一个大的环境下产生的,旨在研究视频监控中的视频测速。
通过查阅文献了解ITS的相关知识和研究了视频及图像处理技术,本文设计基于单目视觉的车辆测速系统。本设计的系统的核心思想是,在对车辆监控的过程中,通过对视频分解成帧图,然后进行图像预处理,能够在保证测速精度的情况下识别出车辆。
测速精度取决于测距的精度。采用单目视觉测距时,传统的方法需要对摄像机的焦距、安装高度和安装角度等参数进行精确标定,方法复杂,而且精度也很难满足要求。本文通过对识别出的车辆用矩形框标记,发现矩形质心与车辆距离镜头呈现一定的关系。通过对实际值和测量值进行分析,然后用最小二乘法进行拟合,得到多项式。将图像中求得的车辆质心代入多项式就可以计算出该车在实际中与镜头的距离。
通过对图片进行预处理算法来消除随机噪声,为了能更精确的识别出车辆本文采用双阈值化消除阴影。采用背景差分的方法,系统能够实现对运动车辆的检测。对检测的车辆用其最小外接矩形框进行标记。
通过实践证明,这种方法相对简单,精度也能满足ITS的要求。
Intelligent Transport System is referred to as ITS. It is a multi-technology system, which includes information technology, data transmission technology, electronic sensor technology, electronic control technology and computer processing technology. It is a wide range, all-round, real-time, more accurate, more efficient integrated transport and management system. China's intelligent transportation system has been used widely in a number of major cities but has not to reach a comprehensive range. Image processing technology develops very fast, which is bound to make a huge boost to the development of the ITS. The thesis is generated in such a large environment, which in order to study video velocimetry in video surveillance.
Upon reviewing of the literature that is related to knowledge of ITS and learning video and image processing technology, this thesis designs a system of measuring vehicle speed based on monocular vision. The core idea of the system is to preprocess the frame factored out from the video and then the system can identify the vehicle in the case of guaranteeing recognition rate and velocity precision.
The accuracy of tachometer depends on the accuracy of the measurement of distance. Monocular vision measuring distance needs to calibrate accurately some parameters in the traditional method, for example the focal length of the camera, installation height and mounting angle. The method is complicated, but the accuracy is also very difficult to meet the requirements. The thesis use rectangles to mark the identified vehicles. The vehicle's external frame mark vehicles identified in this article, which renders a certain relationship between the Y coordinate of the cancroids of the rectangles and the vehicle away from the lens. By measuring the true value and the actual value of these two sets of data, then the thesis use the least squares method to fit unknown relationship, we can obtain connection weights. These connection weights are used as the polynomial coefficients. The distance between the car and the lens can be calculated by substituting the Y coordinate obtained in the image into the polynomial.
The thesis applies preprocessing algorithm on the picture to eliminate random noise, which can get accurate identification of the vehicle. The using of the dual threshold can eliminate shadows, which aims to get accurate identification of the vehicle. According to background difference method, the system can track the vehicle detected. At last with a minimum bounding rectangle of the vehicle detected marks the vehicle detected.
Practice has proved that this method is relatively simple, the accuracy can meet the requirements of the ITS.
通过查阅文献了解ITS的相关知识和研究了视频及图像处理技术,本文设计基于单目视觉的车辆测速系统。本设计的系统的核心思想是,在对车辆监控的过程中,通过对视频分解成帧图,然后进行图像预处理,能够在保证测速精度的情况下识别出车辆。
测速精度取决于测距的精度。采用单目视觉测距时,传统的方法需要对摄像机的焦距、安装高度和安装角度等参数进行精确标定,方法复杂,而且精度也很难满足要求。本文通过对识别出的车辆用矩形框标记,发现矩形质心与车辆距离镜头呈现一定的关系。通过对实际值和测量值进行分析,然后用最小二乘法进行拟合,得到多项式。将图像中求得的车辆质心代入多项式就可以计算出该车在实际中与镜头的距离。
通过对图片进行预处理算法来消除随机噪声,为了能更精确的识别出车辆本文采用双阈值化消除阴影。采用背景差分的方法,系统能够实现对运动车辆的检测。对检测的车辆用其最小外接矩形框进行标记。
通过实践证明,这种方法相对简单,精度也能满足ITS的要求。
Intelligent Transport System is referred to as ITS. It is a multi-technology system, which includes information technology, data transmission technology, electronic sensor technology, electronic control technology and computer processing technology. It is a wide range, all-round, real-time, more accurate, more efficient integrated transport and management system. China's intelligent transportation system has been used widely in a number of major cities but has not to reach a comprehensive range. Image processing technology develops very fast, which is bound to make a huge boost to the development of the ITS. The thesis is generated in such a large environment, which in order to study video velocimetry in video surveillance.
Upon reviewing of the literature that is related to knowledge of ITS and learning video and image processing technology, this thesis designs a system of measuring vehicle speed based on monocular vision. The core idea of the system is to preprocess the frame factored out from the video and then the system can identify the vehicle in the case of guaranteeing recognition rate and velocity precision.
The accuracy of tachometer depends on the accuracy of the measurement of distance. Monocular vision measuring distance needs to calibrate accurately some parameters in the traditional method, for example the focal length of the camera, installation height and mounting angle. The method is complicated, but the accuracy is also very difficult to meet the requirements. The thesis use rectangles to mark the identified vehicles. The vehicle's external frame mark vehicles identified in this article, which renders a certain relationship between the Y coordinate of the cancroids of the rectangles and the vehicle away from the lens. By measuring the true value and the actual value of these two sets of data, then the thesis use the least squares method to fit unknown relationship, we can obtain connection weights. These connection weights are used as the polynomial coefficients. The distance between the car and the lens can be calculated by substituting the Y coordinate obtained in the image into the polynomial.
The thesis applies preprocessing algorithm on the picture to eliminate random noise, which can get accurate identification of the vehicle. The using of the dual threshold can eliminate shadows, which aims to get accurate identification of the vehicle. According to background difference method, the system can track the vehicle detected. At last with a minimum bounding rectangle of the vehicle detected marks the vehicle detected.
Practice has proved that this method is relatively simple, the accuracy can meet the requirements of the ITS.
引文
[1]安全文化网.2012交通事故统计数字[EB].2012.http://aqwh.net.
[2]蔡英凤,张为公,王海.基于视频的城市快速路车辆异常行为检测[J].现代交通技术,2012,1.
[3]鲁丽.基于地感线圈测速仪检定装置的设计[D].[硕士学位论文].郑州：郑州大学,2011.
[4]刘震.基于超声的高速交通流参数获取研究[D].[硕士学位论文].南京：南京理工大学,2006.
[5]徐巍,王桂珠用超声测距法配合PR技术实现交通流特征参数的自动测量[J].自动化与仪表,2000,15(6)：56-59.
[6]王健.视频测速系统中的测距和摄像机标定研究[D].[硕士学位论文].南昌：南昌大学,2008.
[7]李瑞.基于视频的车辆检测与跟踪算法研究[D].[硕士学位论文].上海：上海交通大学,2008.
[8]王荣本,李斌,储江伟,等.公路上基于车载单目机器视觉的前方车距测量方法的研究[J].公路交通科技,2001,18(6)：9-98.
[9]吕乃光,孙鹏,娄小平,韩建栋.结构光三维视觉测量关键技术的研究.[J].自然科学版.2010.
[10]宰金珉,梅国雄.成长曲线在地基沉降预测中的应用.南京建筑工程学院学报,2000,(2)：8-13.
[11]罗昕.实测沉降的拟合和预测[D].杭州：浙江大学,2002.
[12]阎平凡,张长水.人工神经网络与模拟进化计算[M].北京：清华大学出版社.2003.
[13]赵伟.非线性最小二乘法拟合函数[D].[硕士学位论文].吉林：吉林大学,2008.
[14]蒋永馨,王孝通,徐晓刚,等,一种基于光照补偿的图像增强算法[J].电子学报,2009,37(4A)：151-155.
[15]朱虹.数字图像技术与应用[M].机械工业出版社.2011.3.
[16]贾占朝,张亚鸣.基于遗传微粒群混合算法的灰度图形增强[J].计算机技术与发展,2009,17(7)：69-72.
[17]蔚立磊,韩紫恒,张剑.小波变换域图像增强算法及实现[J].信息技术,2009,(9)：132-135,152.
[18]梁晓冰,刘希顺,刘安芝,等.噪声在脉冲耦合神经网络图像增强中的作用[J].中国生物医学工程学报,2009,28(4)：485-489.
[19]贾明桥.大雾天气下的图像清晰化方法[D].西安：西安理工大学,2006.
[20]史贺峰.关于高感光图像去噪算法的研究[D].上海：上海交通大学,2008：54-56.
[21]张小波.基于偏微分方程方法的小波域自适应维纳滤波迭代去噪算法[D].[硕士学位论文],西安：西安电子科技大学,2007.
[22]张瞳,朱虹,张然,等.复小波域维纳滤波与偏微分扩散相结合的图像去噪方法[J].中国图像图形学报,2009,14(5)：848-852.
[23]查宇飞,毕笃彦.基于小波变换的自适应多阈值图像去噪[J].中国图形学报,2005,10(5)：567-570.
[24]张猛,余仲秋,姚绍文.手写体数字识别中图像预处理的研究[J].微计算机信息.2006,22(6)：256-258
[25]李宏,张冬生,林义刚等.基于Otsu理论的灰度图像分割算法研究和改进[J].科学技术与工程,2010,10(22)：5437-5440
[26]谢国庆,白莹,王智文.基于全局迭代阈值和局部分析的护照图像的二值化算法[J].计算机应用与软件,2009,26(11)：118-120
[27]陈丹,张峰,祝贵明.一种改进的文本图像二值化算法[J].计算机工程,2003,29(13)：85-86
[28]童立靖,陈侃,付晓玲等.文档图像二值化算法VFCM[J].计算机工程与应用,2009,30(13)：3216-3218
[29]张继平,刘直芳.背景估计与运动目标检测跟踪[J].计算技术与自动化,2004,51-54.
[30]Baryon J,Heet D.Beachem in S.Performance of optical flow techniques [J]. International Journal of computer Vsion,1994,12(1):42-47.
[31]段修满,孙国霞,基于二次帧差法的智能视频监控系统[C].中国电子学会第十七届信息论学术年会.2010.
[32]万卫兵.智能视频监控中目标检测与识别[M].上海：上海交通大学出版社.2009.11.
[33]王波车辆阴影检测及滤除方法的研究[D].[硕士学位论文].天津：天津大学,2007.
[34]Kenneth R Castleman.数字图像处理[M].朱志刚,等译.北京：电子工业出版社,1998.
[35]李焕.基于视频的车辆测速系统研究[D].[硕士学位论文].郑州：郑州大学,2010.
[36]T.Horprasert,D.Harwood,L.S.Davis.A.statistical.approach for the real time robust background subtraction and shadow detection [C].Proceedings of IEEEI CCV'99 FRAME-RATE Workshop,1999.
[37]PanF,Wang X Y.Wang Q Q.Human detection based on head and shoulder feature. in.intelligent.surveillance.system[J].Journal of Zhe Jiang University(Engineering Science),2004,38(4):397-401.
[38]杨源,楚瀛,毕笃彦.一种基于能量最小化的运动阴影检测方法[J].光电工程,2008.
[39]中华人民共和国公安部.GA 297-2001.机动车测速仪通用技术条件[S].北京：中国标准出版社.2001-3-28.
[2]蔡英凤,张为公,王海.基于视频的城市快速路车辆异常行为检测[J].现代交通技术,2012,1.
[3]鲁丽.基于地感线圈测速仪检定装置的设计[D].[硕士学位论文].郑州：郑州大学,2011.
[4]刘震.基于超声的高速交通流参数获取研究[D].[硕士学位论文].南京：南京理工大学,2006.
[5]徐巍,王桂珠用超声测距法配合PR技术实现交通流特征参数的自动测量[J].自动化与仪表,2000,15(6)：56-59.
[6]王健.视频测速系统中的测距和摄像机标定研究[D].[硕士学位论文].南昌：南昌大学,2008.
[7]李瑞.基于视频的车辆检测与跟踪算法研究[D].[硕士学位论文].上海：上海交通大学,2008.
[8]王荣本,李斌,储江伟,等.公路上基于车载单目机器视觉的前方车距测量方法的研究[J].公路交通科技,2001,18(6)：9-98.
[9]吕乃光,孙鹏,娄小平,韩建栋.结构光三维视觉测量关键技术的研究.[J].自然科学版.2010.
[10]宰金珉,梅国雄.成长曲线在地基沉降预测中的应用.南京建筑工程学院学报,2000,(2)：8-13.
[11]罗昕.实测沉降的拟合和预测[D].杭州：浙江大学,2002.
[12]阎平凡,张长水.人工神经网络与模拟进化计算[M].北京：清华大学出版社.2003.
[13]赵伟.非线性最小二乘法拟合函数[D].[硕士学位论文].吉林：吉林大学,2008.
[14]蒋永馨,王孝通,徐晓刚,等,一种基于光照补偿的图像增强算法[J].电子学报,2009,37(4A)：151-155.
[15]朱虹.数字图像技术与应用[M].机械工业出版社.2011.3.
[16]贾占朝,张亚鸣.基于遗传微粒群混合算法的灰度图形增强[J].计算机技术与发展,2009,17(7)：69-72.
[17]蔚立磊,韩紫恒,张剑.小波变换域图像增强算法及实现[J].信息技术,2009,(9)：132-135,152.
[18]梁晓冰,刘希顺,刘安芝,等.噪声在脉冲耦合神经网络图像增强中的作用[J].中国生物医学工程学报,2009,28(4)：485-489.
[19]贾明桥.大雾天气下的图像清晰化方法[D].西安：西安理工大学,2006.
[20]史贺峰.关于高感光图像去噪算法的研究[D].上海：上海交通大学,2008：54-56.
[21]张小波.基于偏微分方程方法的小波域自适应维纳滤波迭代去噪算法[D].[硕士学位论文],西安：西安电子科技大学,2007.
[22]张瞳,朱虹,张然,等.复小波域维纳滤波与偏微分扩散相结合的图像去噪方法[J].中国图像图形学报,2009,14(5)：848-852.
[23]查宇飞,毕笃彦.基于小波变换的自适应多阈值图像去噪[J].中国图形学报,2005,10(5)：567-570.
[24]张猛,余仲秋,姚绍文.手写体数字识别中图像预处理的研究[J].微计算机信息.2006,22(6)：256-258
[25]李宏,张冬生,林义刚等.基于Otsu理论的灰度图像分割算法研究和改进[J].科学技术与工程,2010,10(22)：5437-5440
[26]谢国庆,白莹,王智文.基于全局迭代阈值和局部分析的护照图像的二值化算法[J].计算机应用与软件,2009,26(11)：118-120
[27]陈丹,张峰,祝贵明.一种改进的文本图像二值化算法[J].计算机工程,2003,29(13)：85-86
[28]童立靖,陈侃,付晓玲等.文档图像二值化算法VFCM[J].计算机工程与应用,2009,30(13)：3216-3218
[29]张继平,刘直芳.背景估计与运动目标检测跟踪[J].计算技术与自动化,2004,51-54.
[30]Baryon J,Heet D.Beachem in S.Performance of optical flow techniques [J]. International Journal of computer Vsion,1994,12(1):42-47.
[31]段修满,孙国霞,基于二次帧差法的智能视频监控系统[C].中国电子学会第十七届信息论学术年会.2010.
[32]万卫兵.智能视频监控中目标检测与识别[M].上海：上海交通大学出版社.2009.11.
[33]王波车辆阴影检测及滤除方法的研究[D].[硕士学位论文].天津：天津大学,2007.
[34]Kenneth R Castleman.数字图像处理[M].朱志刚,等译.北京：电子工业出版社,1998.
[35]李焕.基于视频的车辆测速系统研究[D].[硕士学位论文].郑州：郑州大学,2010.
[36]T.Horprasert,D.Harwood,L.S.Davis.A.statistical.approach for the real time robust background subtraction and shadow detection [C].Proceedings of IEEEI CCV'99 FRAME-RATE Workshop,1999.
[37]PanF,Wang X Y.Wang Q Q.Human detection based on head and shoulder feature. in.intelligent.surveillance.system[J].Journal of Zhe Jiang University(Engineering Science),2004,38(4):397-401.
[38]杨源,楚瀛,毕笃彦.一种基于能量最小化的运动阴影检测方法[J].光电工程,2008.
[39]中华人民共和国公安部.GA 297-2001.机动车测速仪通用技术条件[S].北京：中国标准出版社.2001-3-28.