视频监控下的行人数量统计研究
|
摘要
随着科技的进步,智能视频监控正日益发挥着越来越重要的作用。如今大规模的视频监控系统被广泛应用于各类公共场所,如何对这些海量的监控视频数据进行分析处理进而提取出有用的信息,是目前该领域的一个研究热点和难点。在智能监控的实际应用中,其中有一个关键的问题是对大规模行人数量的统计,目前行人数量的统计不仅仅关系到安防行业,其在交通、商业等方面也有着不可替代的作用,因此对行人数量统计的研究是一个具有重要现实和长远意义的课题。
本文提出一种快慢结合的双高斯背景模型进行运动目标的检测。在该过程中,首先使用K-均值法对背景进行初始化学习并对背景模型的参数赋初值,接着使用背景减除的方法对当前帧进行运动目标的预检测,然后对检测出的候选前景区域进行邻域减除和形态学等一系列的处理,去除那些虚假的前景像素,准确检测出运动目标区域,最后根据检测结果对该背景模型采用快、慢两种更新方式对背景进行维护。除此之外,上述步骤检测出的运动目标区域可能存在阴影,为了能够较为准确的统计出行人数量,阴影的处理必不可少,本文采用基于YCbCr颜色空间模型的算法进行阴影的检测和处理。
本文使用基于底层特征的方法对行人数量进行统计。文中主要选择行人的像素、运动矢量、颜色等特征进行行人数量统计的,该部分具体又可分为在斜角场景和垂直场景下的行人数量统计。在斜角场景下,首先采用上述运动目标检测方法进行前景像素的提取,接着使用LK光流算法进行运动矢量的估计,之后对于设置的虚拟门处的前景像素赋权值,最后进行函数拟合将用到的底层特征转化为能表述行人数量的高层特性,进而对行人数量进行统计。垂直场景具有不必考虑行人间遮挡问题的优点,在垂直场景下,增加头部区域特征的提取,具体来说是提取行人头部发色区域像素,进行如上述斜角场景下的过程对行人数量进行统计。实验表明,本文方法在两种场景下对行人数量的统计准确率能达到90%以上,满足视频监控中行人数量统计的需要。
最后,本文对研究过程中存在的不足进行了分析,并对下一步工作计划做出了阐述。
With the progress of science and technology, the technology of intelligent video surveillance is playing an increasingly important role in our lives day by day. Now large-scale video surveillance systems are widely used in various public places, how to analyze and deal with these massive amounts of surveillance video data to extract the useful information, which is currently a hot research topic and a difficulty in the field of intelligent video surveillance. In the practical application of intelligent monitoring, there is one of the key problems which is the counting of large-scale pedestrians, at present the counting of pedestrians is not only connected with the safety industry, but also plays an irreplaceable role in the traffic, the commerce, et al. Therefore the study of the counting of pedestrians is a subject which has an important practical and long-term significance role.
This paper proposes a kind of fast and slow gaussian mixture model to detect the moving targets. In this step, this paper first use K-means method to initialize the background and the parameters of the background model, and then the background deduction method is used in the current frame for motion target detection, then a series of processes such as the neighborhood deduction, the mathematical morphology, etc, are also used to remove the false foreground pixels in the candidate region which is detected in the above steps, the goal of this process is to accurate the detected moving object regions. At last this paper uses both fast and slow updating ways to maintain the background according to the test results of the background model. In addition, there may be some shadow areas after the above steps when detect the moving targets regions, in order to accurate the statistical number of the pedestrians, the process of shadow removing is necessary, in this paper, the algorithm based on color space model is used for shadow detecting and removing.
This paper uses the method which is based on the underlying characteristics to count the number of pedestrians. This paper mainly choose the pedestrians' foreground pixels, motion vector and color characteristics to count the number. This part can be specific divided into counting the number of pedestrians under the bevel and vertical scene scenarios. In bevel scenario, this paper first uses the moving target detection method which is introduced above for extracting foreground pixel, then uses the optical flow algorithm to estimate the motion vector, and gives weights for the foreground pixels at the virtual door which is set before, finally carries on the function fitting to transform the underlying characteristics into the high-level features which can describe the pedestrians' number. There is an advantage that donnot consider the problem of occlusion In vertical scenario, this paper Increases the extraction of the head region feature, it can be specific to extract pixels of the region of pedestrians' hair, and then use these above steps which are introduced in bevel scenario's to count pedestrians' number. Experiments shows that The accuracy of this method can reach more than90%in both scenarios. So the method of counting pedestrians' number can satisfy the need of video monitoring.
Finally, this paper analyses the shortcomings which are occurred in the course of the research process, and then puts forward the work plan for the next step.
本文提出一种快慢结合的双高斯背景模型进行运动目标的检测。在该过程中,首先使用K-均值法对背景进行初始化学习并对背景模型的参数赋初值,接着使用背景减除的方法对当前帧进行运动目标的预检测,然后对检测出的候选前景区域进行邻域减除和形态学等一系列的处理,去除那些虚假的前景像素,准确检测出运动目标区域,最后根据检测结果对该背景模型采用快、慢两种更新方式对背景进行维护。除此之外,上述步骤检测出的运动目标区域可能存在阴影,为了能够较为准确的统计出行人数量,阴影的处理必不可少,本文采用基于YCbCr颜色空间模型的算法进行阴影的检测和处理。
本文使用基于底层特征的方法对行人数量进行统计。文中主要选择行人的像素、运动矢量、颜色等特征进行行人数量统计的,该部分具体又可分为在斜角场景和垂直场景下的行人数量统计。在斜角场景下,首先采用上述运动目标检测方法进行前景像素的提取,接着使用LK光流算法进行运动矢量的估计,之后对于设置的虚拟门处的前景像素赋权值,最后进行函数拟合将用到的底层特征转化为能表述行人数量的高层特性,进而对行人数量进行统计。垂直场景具有不必考虑行人间遮挡问题的优点,在垂直场景下,增加头部区域特征的提取,具体来说是提取行人头部发色区域像素,进行如上述斜角场景下的过程对行人数量进行统计。实验表明,本文方法在两种场景下对行人数量的统计准确率能达到90%以上,满足视频监控中行人数量统计的需要。
最后,本文对研究过程中存在的不足进行了分析,并对下一步工作计划做出了阐述。
With the progress of science and technology, the technology of intelligent video surveillance is playing an increasingly important role in our lives day by day. Now large-scale video surveillance systems are widely used in various public places, how to analyze and deal with these massive amounts of surveillance video data to extract the useful information, which is currently a hot research topic and a difficulty in the field of intelligent video surveillance. In the practical application of intelligent monitoring, there is one of the key problems which is the counting of large-scale pedestrians, at present the counting of pedestrians is not only connected with the safety industry, but also plays an irreplaceable role in the traffic, the commerce, et al. Therefore the study of the counting of pedestrians is a subject which has an important practical and long-term significance role.
This paper proposes a kind of fast and slow gaussian mixture model to detect the moving targets. In this step, this paper first use K-means method to initialize the background and the parameters of the background model, and then the background deduction method is used in the current frame for motion target detection, then a series of processes such as the neighborhood deduction, the mathematical morphology, etc, are also used to remove the false foreground pixels in the candidate region which is detected in the above steps, the goal of this process is to accurate the detected moving object regions. At last this paper uses both fast and slow updating ways to maintain the background according to the test results of the background model. In addition, there may be some shadow areas after the above steps when detect the moving targets regions, in order to accurate the statistical number of the pedestrians, the process of shadow removing is necessary, in this paper, the algorithm based on color space model is used for shadow detecting and removing.
This paper uses the method which is based on the underlying characteristics to count the number of pedestrians. This paper mainly choose the pedestrians' foreground pixels, motion vector and color characteristics to count the number. This part can be specific divided into counting the number of pedestrians under the bevel and vertical scene scenarios. In bevel scenario, this paper first uses the moving target detection method which is introduced above for extracting foreground pixel, then uses the optical flow algorithm to estimate the motion vector, and gives weights for the foreground pixels at the virtual door which is set before, finally carries on the function fitting to transform the underlying characteristics into the high-level features which can describe the pedestrians' number. There is an advantage that donnot consider the problem of occlusion In vertical scenario, this paper Increases the extraction of the head region feature, it can be specific to extract pixels of the region of pedestrians' hair, and then use these above steps which are introduced in bevel scenario's to count pedestrians' number. Experiments shows that The accuracy of this method can reach more than90%in both scenarios. So the method of counting pedestrians' number can satisfy the need of video monitoring.
Finally, this paper analyses the shortcomings which are occurred in the course of the research process, and then puts forward the work plan for the next step.
引文
[1]Benabbas Y, Ihaddadene N, Yahiaoui T, et al. Spatio-Temporal Optical Flow Analysis for People Counting[C]. In:Proceedings of the 7th IEEE International Conference on Advanced Video and Signal Based Surveillance. Boston:Advanced Video and Signal Based Surveillance,2010,212-217.
[2]张卿.基于视频的人流检测与分析研究[D].上海交通大学,2012.
[3]Thou-Ho Chen, Tsong-Yi Chen, Zhi-Xian Chen. An Intelligent People-Flow Counting Method for Passing Through a Gate[C]. In:Proceedings of the IEEE Conference on Robotics, Automation and Mechatronics. Bangkok:Robotics, Automation and Mechatronics, 2006,1-6.
[4]Terada K, Yoshida D, Oe S, Yamaguchi J. A method of counting the passing people by using the stereo images[C]. In:Proceedings of the IEEE International Conference on Image Processing, Kobe:Image Processing,1999,338-342.
[5]郑翔翔,张怡.基于头部检测和跟踪的人数统计方法[J].计算机应用与软件,2011,(02)：44-46+102.
[6]Xiao-Wei Xu, Zhi-Yan Wang, Ying-Hong Liang, et al. A Rapid Method for Passing People Counting in Monocular Video Sequences[C]. In:Proceedings of the 6th International Conference on Machine Learning and Cybernetics. Hong Kong:Machine Learning and Cybernetics,2007,1657-1662.
[7]Kilambi P, Ribnick E, Joshi A.J, et al. Estimating pedestrian counts in groups[J]. Computer Vision and Image Understanding,2008,110:43-59.
[8]Gwang-Gook Lee, Byeoung-su Kim, Whoi-Yul Kim. Automatic Estimation of Pedestrian Flow[C]. In:Proceedings of the 1st ACM/IEEE International Conference on Distributed Smart Cameras. Vienna:Distributed Smart Cameras,2007,291-296.
[9]Jiajun Lu, Yi Xu, Xiaokang Yang. Counting pedestrians and cars with an improved virtual gate method[C]. In:Proceedings of the International Conference on Computer Application and System Modeling, Proceedings. Shanxi:Computer Application and System Modeling, 2010,448-452.
[10]汤石晨,陈锻生.基于光流法和行人面积特征的人数统计方法研究[J].图学学报,2013,(01)：139-144.
[11]姚金良,钱翰博,汪澄.视觉自适应行人计数[J].光电工程,2012,(07)：102-108.
[12]Ya-Li Hou, Pang G.K.-H.. Automated people counting at a mass site[C]. In:Proceedings of the IEEE International Conference on Automation and Logistics. Qingdao:Automation and Logistics,2008,464-469.
[13]姜兰池,沈国强,张国煊.基于HSV分块颜色直方图的图像检索算法[J].机电工程,2009,(11)：54-57.
[14]陈燕萍.基于背景减除的运动目标检测算法研究[D].厦门大学,2008.
[15]Barnich O, Van Droogenbroeck M. ViBe:A Universal Background Subtraction Algorithm for Video Sequences[J]. IEEE Transactions on Image Processing,2011,20:1709-1724.
[16]李毅,孙正兴,远博,张岩.一种改进的帧差和背景减相结合的运动检测方法[J].中国图象图形学报,2009,(06)：1162-1168.
[17]李松.基于视频序列的人物跟踪与行为分析技术研究[D].哈尔滨工程大学,2010.
[18]Duan-Yu Chen, Chih-Wen Su, Yi-Chong Zeng,et al. An online people counting system for electronic advertising machines[C]. In:Proceedings of the IEEE International Conference on Multimedia and Expo. New York:Multimedia and Expo,2009,1262-1265.
[19]Barandiaran J, Murguia B, Boto F. Real-Time People Counting Using Multiple Lines[C]. In: Proceedings of the 9th International Workshop on Image Analysis for Multimedia Interactive Services. Klagenfurt:Image Analysis for Multimedia Interactive Services,2008,159-162.
[20]陈静.图像配准特征点提取算法研究[D].南京理工大学,2006.
[21]张勇.基于主动轮廓的运动目标检测与跟踪[D].武汉理工大学,2010.
[22]Shengsheng Yu, Xiaoping Chen, Weiping Sun, et al. A robust method for detecting and counting people[C]. In:Proceedings of the International Conference on Audio, Language and Image Processing, Proceedings. Shanghai:Audio, Language and Image Processing, 2008,1545-1549.
[23]Antic B, Letic D, Culibrk D, et al. K-means based segmentation for real-time zenithal people counting[C]. In:Proceedings of the International Conference on Image Processing. Cairo: Image Processing,2009,2565-2568.
[24]Yang Cong, Haifeng Gong, Song-Chun Zhu, et al. Flow mosaicking:Real-time pedestrian counting without scene-specific learning[C]. In:Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops. Miami:Computer Vision and Pattern Recognition,2009,1093-1100.
[25]Byeoung-Su K, Gwang-Gook L, Ja-Young Y. A method of counting pedestrians in crowded scenes[C]. In:Proceedings of the 4th International Conference on Intelligent Computing. Shanghai:Advanced Intelligent Computing Theories and Applications:With Aspects of Artificial Intelligence,2008,1117-1126.
[26]黄斯茜,李光.智能视频监控系统运动目标检测技术研究综述[J].信息通信,2012,(04)：57-58.
[27]查成东,王长松,巩宪锋,周家新.基于自适应背景模型的运动目标检测[J].光电工程,2008,(01)：26-30.
[28]齐美彬,王倩,蒋建国,安宝磊.基于背景像素值频次最高假设的背景重构算法[J].中国图象图形学报,2008,(04)：723-728.
[29]张恒,胡文龙,丁赤飚.一种自适应学习的混合高斯模型视频目标检测算法[J].中国图象图形学报,2010,(04)：631-636.
[30]Yunyi Liu, Zhiyan Bin, The Improved Moving Object Detection and Shadow Removing Algorithms for Video Surveillance[C]. In:Proceedings of the International Conference on Computational Intelligence and Software Engineering. Wuhan:Computational Intelligence and Software Engineering,2010,1-5.
[31]Zhenhong Shang, Jinxia Shang, Xiaofang Cai, et al. Shadow removing based on color and texture features[C]. In:Proceedings of the International Conference on Consumer Electronics, Communications and Networks. XianNing:Consumer Electronics, Communications and Networks,2011,2578-2581.
[32]Salamati N, Germain A, Susstrunk S. Removing shadows from images using color and near-infrared[C]. In:Proceedings of the 18th IEEE International Conference on Image Processing,2011,1713-1716.
[33]付萍,方帅,徐心和,薛定宇.视频监控系统中运动目标检测的阴影去除方法[J].计算机工程,2007,(10)：22-24.
[34]邹波,管业鹏,吴进.基于颜色信息和光学增益的阴影消除[J].计算机仿真,2009,(10)：261-264.
[35]邓亚丽,毋立芳,李云腾.一种有效的图像阴影自动去除算法[J].信号处理,2011,(11)：1724-1728.
[36]王勇.基于统计方法的运动目标检测与跟踪技术研究[D].华中科技大学,2009.
[37]Lukas B, Kanade T. An Iterative Image Registration Technique with an Application to Stereo Vision[C]. In:Proceedings of the 7th International Joint Conference on Artificial Intelligence. 1981,674-679.
[38]Stauffer C, Grimson W. E. L. Adaptive background mixture models for real-time tracking[J]. In:Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition. Fort Collin:Computer Vision and Pattern Recognition,1999,02:246-252
[39]赵敏.单目视觉多行人目标检测与跟踪技术研究[D].重庆大学,2010,22-25.
[40]Lee G.G, Song S. H, Kim W-Y. Crowd Density Estimation for Video Surveillance. In: Proceedings of the International Conference on Computing, Communications and Control Technologies,2007 196-199.
[41]Kim Z. Real Time Object Tracking based on Dynamic Feature Grouping with Background Subtraction[C]. In:Proceedings of the 26th IEEE Conference on Computer Vision and Pattern Recognition,Anchorage:Computer Vision and Pattern Recognition,2008.
[2]张卿.基于视频的人流检测与分析研究[D].上海交通大学,2012.
[3]Thou-Ho Chen, Tsong-Yi Chen, Zhi-Xian Chen. An Intelligent People-Flow Counting Method for Passing Through a Gate[C]. In:Proceedings of the IEEE Conference on Robotics, Automation and Mechatronics. Bangkok:Robotics, Automation and Mechatronics, 2006,1-6.
[4]Terada K, Yoshida D, Oe S, Yamaguchi J. A method of counting the passing people by using the stereo images[C]. In:Proceedings of the IEEE International Conference on Image Processing, Kobe:Image Processing,1999,338-342.
[5]郑翔翔,张怡.基于头部检测和跟踪的人数统计方法[J].计算机应用与软件,2011,(02)：44-46+102.
[6]Xiao-Wei Xu, Zhi-Yan Wang, Ying-Hong Liang, et al. A Rapid Method for Passing People Counting in Monocular Video Sequences[C]. In:Proceedings of the 6th International Conference on Machine Learning and Cybernetics. Hong Kong:Machine Learning and Cybernetics,2007,1657-1662.
[7]Kilambi P, Ribnick E, Joshi A.J, et al. Estimating pedestrian counts in groups[J]. Computer Vision and Image Understanding,2008,110:43-59.
[8]Gwang-Gook Lee, Byeoung-su Kim, Whoi-Yul Kim. Automatic Estimation of Pedestrian Flow[C]. In:Proceedings of the 1st ACM/IEEE International Conference on Distributed Smart Cameras. Vienna:Distributed Smart Cameras,2007,291-296.
[9]Jiajun Lu, Yi Xu, Xiaokang Yang. Counting pedestrians and cars with an improved virtual gate method[C]. In:Proceedings of the International Conference on Computer Application and System Modeling, Proceedings. Shanxi:Computer Application and System Modeling, 2010,448-452.
[10]汤石晨,陈锻生.基于光流法和行人面积特征的人数统计方法研究[J].图学学报,2013,(01)：139-144.
[11]姚金良,钱翰博,汪澄.视觉自适应行人计数[J].光电工程,2012,(07)：102-108.
[12]Ya-Li Hou, Pang G.K.-H.. Automated people counting at a mass site[C]. In:Proceedings of the IEEE International Conference on Automation and Logistics. Qingdao:Automation and Logistics,2008,464-469.
[13]姜兰池,沈国强,张国煊.基于HSV分块颜色直方图的图像检索算法[J].机电工程,2009,(11)：54-57.
[14]陈燕萍.基于背景减除的运动目标检测算法研究[D].厦门大学,2008.
[15]Barnich O, Van Droogenbroeck M. ViBe:A Universal Background Subtraction Algorithm for Video Sequences[J]. IEEE Transactions on Image Processing,2011,20:1709-1724.
[16]李毅,孙正兴,远博,张岩.一种改进的帧差和背景减相结合的运动检测方法[J].中国图象图形学报,2009,(06)：1162-1168.
[17]李松.基于视频序列的人物跟踪与行为分析技术研究[D].哈尔滨工程大学,2010.
[18]Duan-Yu Chen, Chih-Wen Su, Yi-Chong Zeng,et al. An online people counting system for electronic advertising machines[C]. In:Proceedings of the IEEE International Conference on Multimedia and Expo. New York:Multimedia and Expo,2009,1262-1265.
[19]Barandiaran J, Murguia B, Boto F. Real-Time People Counting Using Multiple Lines[C]. In: Proceedings of the 9th International Workshop on Image Analysis for Multimedia Interactive Services. Klagenfurt:Image Analysis for Multimedia Interactive Services,2008,159-162.
[20]陈静.图像配准特征点提取算法研究[D].南京理工大学,2006.
[21]张勇.基于主动轮廓的运动目标检测与跟踪[D].武汉理工大学,2010.
[22]Shengsheng Yu, Xiaoping Chen, Weiping Sun, et al. A robust method for detecting and counting people[C]. In:Proceedings of the International Conference on Audio, Language and Image Processing, Proceedings. Shanghai:Audio, Language and Image Processing, 2008,1545-1549.
[23]Antic B, Letic D, Culibrk D, et al. K-means based segmentation for real-time zenithal people counting[C]. In:Proceedings of the International Conference on Image Processing. Cairo: Image Processing,2009,2565-2568.
[24]Yang Cong, Haifeng Gong, Song-Chun Zhu, et al. Flow mosaicking:Real-time pedestrian counting without scene-specific learning[C]. In:Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops. Miami:Computer Vision and Pattern Recognition,2009,1093-1100.
[25]Byeoung-Su K, Gwang-Gook L, Ja-Young Y. A method of counting pedestrians in crowded scenes[C]. In:Proceedings of the 4th International Conference on Intelligent Computing. Shanghai:Advanced Intelligent Computing Theories and Applications:With Aspects of Artificial Intelligence,2008,1117-1126.
[26]黄斯茜,李光.智能视频监控系统运动目标检测技术研究综述[J].信息通信,2012,(04)：57-58.
[27]查成东,王长松,巩宪锋,周家新.基于自适应背景模型的运动目标检测[J].光电工程,2008,(01)：26-30.
[28]齐美彬,王倩,蒋建国,安宝磊.基于背景像素值频次最高假设的背景重构算法[J].中国图象图形学报,2008,(04)：723-728.
[29]张恒,胡文龙,丁赤飚.一种自适应学习的混合高斯模型视频目标检测算法[J].中国图象图形学报,2010,(04)：631-636.
[30]Yunyi Liu, Zhiyan Bin, The Improved Moving Object Detection and Shadow Removing Algorithms for Video Surveillance[C]. In:Proceedings of the International Conference on Computational Intelligence and Software Engineering. Wuhan:Computational Intelligence and Software Engineering,2010,1-5.
[31]Zhenhong Shang, Jinxia Shang, Xiaofang Cai, et al. Shadow removing based on color and texture features[C]. In:Proceedings of the International Conference on Consumer Electronics, Communications and Networks. XianNing:Consumer Electronics, Communications and Networks,2011,2578-2581.
[32]Salamati N, Germain A, Susstrunk S. Removing shadows from images using color and near-infrared[C]. In:Proceedings of the 18th IEEE International Conference on Image Processing,2011,1713-1716.
[33]付萍,方帅,徐心和,薛定宇.视频监控系统中运动目标检测的阴影去除方法[J].计算机工程,2007,(10)：22-24.
[34]邹波,管业鹏,吴进.基于颜色信息和光学增益的阴影消除[J].计算机仿真,2009,(10)：261-264.
[35]邓亚丽,毋立芳,李云腾.一种有效的图像阴影自动去除算法[J].信号处理,2011,(11)：1724-1728.
[36]王勇.基于统计方法的运动目标检测与跟踪技术研究[D].华中科技大学,2009.
[37]Lukas B, Kanade T. An Iterative Image Registration Technique with an Application to Stereo Vision[C]. In:Proceedings of the 7th International Joint Conference on Artificial Intelligence. 1981,674-679.
[38]Stauffer C, Grimson W. E. L. Adaptive background mixture models for real-time tracking[J]. In:Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition. Fort Collin:Computer Vision and Pattern Recognition,1999,02:246-252
[39]赵敏.单目视觉多行人目标检测与跟踪技术研究[D].重庆大学,2010,22-25.
[40]Lee G.G, Song S. H, Kim W-Y. Crowd Density Estimation for Video Surveillance. In: Proceedings of the International Conference on Computing, Communications and Control Technologies,2007 196-199.
[41]Kim Z. Real Time Object Tracking based on Dynamic Feature Grouping with Background Subtraction[C]. In:Proceedings of the 26th IEEE Conference on Computer Vision and Pattern Recognition,Anchorage:Computer Vision and Pattern Recognition,2008.