视频分析法在公交车客流统计中的研究与应用
|
摘要
客流量的自动统计是智能公共交通系统的重要组成部分,通过实时获得各个站点的上下车人数,可以分析得出客流在线路、方向、数量等方面的分布规律,为公共交通工具的所有者和管理者提供可靠的数据支持,从而提高系统的运行效率,有效缓解日益沉重的交通负担。
针对国内公共交通环境,总结了现有的几种客流量自动统计的实现方法,对比了几种方法的优缺点,考虑到客流信息的完整性是获得高准确率的前提,选择视频分析法作为研究内容。以“俯视”角度录制的乘客上下车视频作为实验数据,针对每一帧图像,首先进行平滑、去噪等预处理,然后针对目标在不同方面的特征,尝试采用改进的Hough变换、区域生长法、背景减除法三种方法分别进行乘客目标的分割,之后再对相邻两帧图像的目标进行匹配、识别,从而实现各个乘客的跟踪、统计。
为简化情况验证算法的优劣,以自行录制的行人进出视频来模拟公交客流的上下车、目标拥挤、静止等情况,通过大量的实验结果表明视频分析法相对其他的几种统计方法能达到更高的准确率,尤其在人多拥挤的情况下,其他几种方法往往因为遮挡不能获得足够完整的目标信息。同时可以看出目标的有效分割是提高视频统计准确率的关键,其中的Hough变换对目标形状有“近似圆形”的要求,区域生长法要求目标的灰度值集中在较小的范围,而背景减除法对目标的要求很少,并且通过改进混合高斯背景模型取得了较好的效果。
Auto passenger count system is an important part of APTS(Advanced Public Transportation Systems), which can in real time get the count of people getting on or getting off at every bus stop, and then conclude the passenger flow’s distributing rules for lines, directions, amount etc. It provides reliable data support for the owner and manager of public transports, so they can improve the efficiency of system, effectively lower the burden of public traffic.
In view of the conditions of civil public transport, summarized the existing ways of auto passenger counting, contrasted their advantages and disadvantages. Considering the integrity of passenger flow information is premise of improving accuracy, chose video analysis as study subject. Videos recorded by a vertical camera are experimental data. Every frame was first pretreated by smoothing noise, then according to the different characteristics of objects, tried to segment the objects using Hough-transform, region-growing, and background-subtraction respectively, next, matched and recognized every object in two adjacent frames, finally implemented the tacking and statistics of every passenger.
In order to contrast different methods, simplified the conditions by recording videos of walker follow to simulate getting-on, getting-off, crowd and stillness of the bus passenger flow. Many experiments showed that video analysis provides passenger flow counting with higher accuracy, especially in the rush-hour, other ways often can not get complete information. On the other hand, effective segmentation of objects is crucial to improve accuracy of counting. Hough-transform is used to detect objects like a circle; region-growing needs objects’gray value to be in narrow limits; background-subtraction is not strict with objects and achieved good results by improving Gauss background model.
针对国内公共交通环境,总结了现有的几种客流量自动统计的实现方法,对比了几种方法的优缺点,考虑到客流信息的完整性是获得高准确率的前提,选择视频分析法作为研究内容。以“俯视”角度录制的乘客上下车视频作为实验数据,针对每一帧图像,首先进行平滑、去噪等预处理,然后针对目标在不同方面的特征,尝试采用改进的Hough变换、区域生长法、背景减除法三种方法分别进行乘客目标的分割,之后再对相邻两帧图像的目标进行匹配、识别,从而实现各个乘客的跟踪、统计。
为简化情况验证算法的优劣,以自行录制的行人进出视频来模拟公交客流的上下车、目标拥挤、静止等情况,通过大量的实验结果表明视频分析法相对其他的几种统计方法能达到更高的准确率,尤其在人多拥挤的情况下,其他几种方法往往因为遮挡不能获得足够完整的目标信息。同时可以看出目标的有效分割是提高视频统计准确率的关键,其中的Hough变换对目标形状有“近似圆形”的要求,区域生长法要求目标的灰度值集中在较小的范围,而背景减除法对目标的要求很少,并且通过改进混合高斯背景模型取得了较好的效果。
Auto passenger count system is an important part of APTS(Advanced Public Transportation Systems), which can in real time get the count of people getting on or getting off at every bus stop, and then conclude the passenger flow’s distributing rules for lines, directions, amount etc. It provides reliable data support for the owner and manager of public transports, so they can improve the efficiency of system, effectively lower the burden of public traffic.
In view of the conditions of civil public transport, summarized the existing ways of auto passenger counting, contrasted their advantages and disadvantages. Considering the integrity of passenger flow information is premise of improving accuracy, chose video analysis as study subject. Videos recorded by a vertical camera are experimental data. Every frame was first pretreated by smoothing noise, then according to the different characteristics of objects, tried to segment the objects using Hough-transform, region-growing, and background-subtraction respectively, next, matched and recognized every object in two adjacent frames, finally implemented the tacking and statistics of every passenger.
In order to contrast different methods, simplified the conditions by recording videos of walker follow to simulate getting-on, getting-off, crowd and stillness of the bus passenger flow. Many experiments showed that video analysis provides passenger flow counting with higher accuracy, especially in the rush-hour, other ways often can not get complete information. On the other hand, effective segmentation of objects is crucial to improve accuracy of counting. Hough-transform is used to detect objects like a circle; region-growing needs objects’gray value to be in narrow limits; background-subtraction is not strict with objects and achieved good results by improving Gauss background model.
引文
[1]王国良.图像处理技术在智能交通系统中的应用研究[D].大连:大连海事大学,2007.
[2]汪东.信息技术在现代汽车及交通领域中的应用[J].交通世界,2005,21(8):16-18.
[3]田冠华.公交客流量监测车载终端的设计与实现[D].大连:大连理工大学,2006.
[4]李志刚.基于图像的公共汽车人数自动统计技术研究[D].太原:中北大学,2008.
[5]赵敏.基于多运动目标识别的自动乘客计数技术研究[D].重庆:重庆大学,2006.
[6]屈建中,张军,鲁墨武.长途客车流量监测装置[J].测控技术,1994,3(13).
[7]衣淑凤.基于小波变换和神经网络的人数统计方法[D].北京:北京工业大学,2004.
[8]薛亮,吴洋.公交客流统计中的目标检测研究[J].人工智能及识别技术,2008,20(3):1.
[9]王聪.基于机器视觉的客流统计方法与实现[D].重庆:西南大学,2009.
[10]王晓,魏志强,周利江,等.基于脚印方向识别的客流检测与监控系统[J].城市交通,2005,8(3):3.
[11] Ezaki,H. Nanbu,T. Nincmiya,et al.Estination of liquidus temperature of Sn-based alloys and its application to the design of Pb-free solder[J].Journal of Materials Science:Materials in Electronics,2002,13(5):269-272.
[12] Mae Y,Shirai Y,Miura J,et al. Object tracking in clutter background based on optical flow and edges[J].Proceedings of the 13th International Conference on Pattern Recognition,1996,1(25):196-200.
[13]吕沛.智能视频监控中的行人流量自动统计技术的研究[D].武汉:武汉理工大学,2010.
[14]杨岚.人体目标识别及其在客流量统计中的应用[D].大连:大连海事大学,2006.
[15]巨朋真.基于头肩特征的人体目标检测与跟踪[D].大连:大连海事大学,2008.
[16]胡迎春,张增芳,邓翔.人脸图像特征点眼睛的定位与提取[J].广西工学院学报,2003,14(1):35-39.
[17]潘峰,王宣银,王全强.智能监控中基于头肩特征的人体检测方法研究[J].浙江大学学报(工学版) ,2004,38(4):397-401.
[18]韩鸿哲,李彬,王志良,等.基于傅立叶描述子的步态识别[J].计算机工程,2005,31(2):48-50.
[19] Milan Sonka,Vaclav Hlavac,Roger Boyle.Image processing,Analysis and Machine Vision.3rd Edition[M].Florence:Thomson Learning,1999.163-172.
[20] Chantal Revol-Muller,Francoise Peyrin,Yannick Carrillon, et al. Automated 3D region growing algorithm based on an assessmen function[J].Pattern Recognition Letters,2002(3):137-150.
[21] R Adams,L Bischof. Seeded region growing[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,1994,16(6): 641-647.
[22]李政国.基于区域生长法的高空间分辨率遥感图像分割与实现[D].广西南宁:广西大学,2008.
[23]章毓晋.图像分割[M].北京:科学出版社,2001.
[24] Frank Y, Shih, Shouxian Cheng.Automatic seeded region growing for color image segmentation[J]. Image and Vision Computing,2005(23): 877-886.
[25]邵文坤,黄爱民,韦庆.目标跟踪方法综述[J].影像技术,2006, 3(2):58-64.
[26] F.M.porikli,O.Tuzel.Human Body Tracking by Adaptive Background Models and Mean-shift Analysis[J].IEEE International Workshop on Performance Evaluation of Tracking and Surveillance,2003:15-21.
[27] B.Scholkopf.Statistical Learning and Kernel Methods[J].Technology Report MSR-T,2002:101-107.
[28] B.H.Lee,II Choi,GJ.Jeon.Motion-based Moving Object Tracking Using an ActiveContour[J].Acoustics,Speech and Signal Processing,2006:223-229.
[29] H.K.W Kim,J.D.Kim.Region-based Shape Descriptor Invariant to Rotation,Scale and Translation[J].Signal Processing,2003:189-197.
[30]田军.动态背景下运动目标检测与跟踪的研究[D].北京:华北电力大学,2009.
[31] Yang Wei, Zhang Tianwen.A new method for the detection of moving targets in complex scenes[J].Computer Research & Development,1998,35(8):724-728.
[32] Wu Zeju,Chen Jundong,Liu Yun,et al.Video object segmentation of still background[J].Journal of Qingdao University of Science and Technology,2004,25(5):457-460.
[33]于成忠,朱骏,袁晓辉.基于背景差法的运动目标检测[J].东南大学学报(自然科学版),2005,35(2):159-160.
[34]侯志强,韩崇昭.基于像素灰度归类的背景重构算法[J].软件学报, 2005, 9(16).
[35] Long W,Yang Y.Stationary background generation:An alternative to the difference of two images[J].Pattern Recognition,1990,23(12):1351-1359.
[36] Gloyer B,Aghajan HK,Siu KY,et al.Video-Based freeway monitoring system using recursive vehicle tracking[J].In:Proc.ofthe IS&T-SPIE Symp.on Electronic Imaging:Image and Video Processing, 1995,9(2421):173-180.
[37] Haritaoglu I,Harwood D,Davis L S.W4:Rea1-time Surveillance of People and Their Activities[J].IEEE Transaction on Pattern Analysis and Machine Intelligence,2000,22(8).
[38]钱晋.基于背景差法的视频车辆检测算法研究[D].上海:上海交通大学,2007.
[39]李扬.视频检测中高斯分布模型的应用[D].长春:吉林大学,2007.
[40]谭文明.复杂背景建模方法及其在运动目标检测中的应用[D].长沙:国防科学技术大学,2009.
[41] Stauffer C,WEL Crimson.Adaptive background mixture models for real-time tracking[C].Proc of the IEEE Conference on Computer Vision and Pattern Recognition,1999,2:246-252.
[42] Power P W,Schoonees J A.Understanding background mixture models for foregrounds segmentation[C].In: Proceedings of Image and Vision Computing. New Zealand: Auckland,2002,267-271.
[43] Nir Friedman,Stuart Russell.Image segmentation in video sequences:A probabilistic approach[C].Proc.of the 13th Conf on Uncertainty in Artificial Intelligence,1997,97:175-181.
[44] M.Piccardi.Background Subtraction Techniques:A Review[C].2004 IEEE International Conference on Systems,Man and Cybernetics,2004,3099-3104.
[45] C.Stauffer,W.Eric,L.Grimson.Learning Patterns of Activity Using Real-Time Tracking[J].IEEE Transactions on Pattern Analysis and Machine Intelligence, 2000, 8, (22):747-757.
[2]汪东.信息技术在现代汽车及交通领域中的应用[J].交通世界,2005,21(8):16-18.
[3]田冠华.公交客流量监测车载终端的设计与实现[D].大连:大连理工大学,2006.
[4]李志刚.基于图像的公共汽车人数自动统计技术研究[D].太原:中北大学,2008.
[5]赵敏.基于多运动目标识别的自动乘客计数技术研究[D].重庆:重庆大学,2006.
[6]屈建中,张军,鲁墨武.长途客车流量监测装置[J].测控技术,1994,3(13).
[7]衣淑凤.基于小波变换和神经网络的人数统计方法[D].北京:北京工业大学,2004.
[8]薛亮,吴洋.公交客流统计中的目标检测研究[J].人工智能及识别技术,2008,20(3):1.
[9]王聪.基于机器视觉的客流统计方法与实现[D].重庆:西南大学,2009.
[10]王晓,魏志强,周利江,等.基于脚印方向识别的客流检测与监控系统[J].城市交通,2005,8(3):3.
[11] Ezaki,H. Nanbu,T. Nincmiya,et al.Estination of liquidus temperature of Sn-based alloys and its application to the design of Pb-free solder[J].Journal of Materials Science:Materials in Electronics,2002,13(5):269-272.
[12] Mae Y,Shirai Y,Miura J,et al. Object tracking in clutter background based on optical flow and edges[J].Proceedings of the 13th International Conference on Pattern Recognition,1996,1(25):196-200.
[13]吕沛.智能视频监控中的行人流量自动统计技术的研究[D].武汉:武汉理工大学,2010.
[14]杨岚.人体目标识别及其在客流量统计中的应用[D].大连:大连海事大学,2006.
[15]巨朋真.基于头肩特征的人体目标检测与跟踪[D].大连:大连海事大学,2008.
[16]胡迎春,张增芳,邓翔.人脸图像特征点眼睛的定位与提取[J].广西工学院学报,2003,14(1):35-39.
[17]潘峰,王宣银,王全强.智能监控中基于头肩特征的人体检测方法研究[J].浙江大学学报(工学版) ,2004,38(4):397-401.
[18]韩鸿哲,李彬,王志良,等.基于傅立叶描述子的步态识别[J].计算机工程,2005,31(2):48-50.
[19] Milan Sonka,Vaclav Hlavac,Roger Boyle.Image processing,Analysis and Machine Vision.3rd Edition[M].Florence:Thomson Learning,1999.163-172.
[20] Chantal Revol-Muller,Francoise Peyrin,Yannick Carrillon, et al. Automated 3D region growing algorithm based on an assessmen function[J].Pattern Recognition Letters,2002(3):137-150.
[21] R Adams,L Bischof. Seeded region growing[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,1994,16(6): 641-647.
[22]李政国.基于区域生长法的高空间分辨率遥感图像分割与实现[D].广西南宁:广西大学,2008.
[23]章毓晋.图像分割[M].北京:科学出版社,2001.
[24] Frank Y, Shih, Shouxian Cheng.Automatic seeded region growing for color image segmentation[J]. Image and Vision Computing,2005(23): 877-886.
[25]邵文坤,黄爱民,韦庆.目标跟踪方法综述[J].影像技术,2006, 3(2):58-64.
[26] F.M.porikli,O.Tuzel.Human Body Tracking by Adaptive Background Models and Mean-shift Analysis[J].IEEE International Workshop on Performance Evaluation of Tracking and Surveillance,2003:15-21.
[27] B.Scholkopf.Statistical Learning and Kernel Methods[J].Technology Report MSR-T,2002:101-107.
[28] B.H.Lee,II Choi,GJ.Jeon.Motion-based Moving Object Tracking Using an ActiveContour[J].Acoustics,Speech and Signal Processing,2006:223-229.
[29] H.K.W Kim,J.D.Kim.Region-based Shape Descriptor Invariant to Rotation,Scale and Translation[J].Signal Processing,2003:189-197.
[30]田军.动态背景下运动目标检测与跟踪的研究[D].北京:华北电力大学,2009.
[31] Yang Wei, Zhang Tianwen.A new method for the detection of moving targets in complex scenes[J].Computer Research & Development,1998,35(8):724-728.
[32] Wu Zeju,Chen Jundong,Liu Yun,et al.Video object segmentation of still background[J].Journal of Qingdao University of Science and Technology,2004,25(5):457-460.
[33]于成忠,朱骏,袁晓辉.基于背景差法的运动目标检测[J].东南大学学报(自然科学版),2005,35(2):159-160.
[34]侯志强,韩崇昭.基于像素灰度归类的背景重构算法[J].软件学报, 2005, 9(16).
[35] Long W,Yang Y.Stationary background generation:An alternative to the difference of two images[J].Pattern Recognition,1990,23(12):1351-1359.
[36] Gloyer B,Aghajan HK,Siu KY,et al.Video-Based freeway monitoring system using recursive vehicle tracking[J].In:Proc.ofthe IS&T-SPIE Symp.on Electronic Imaging:Image and Video Processing, 1995,9(2421):173-180.
[37] Haritaoglu I,Harwood D,Davis L S.W4:Rea1-time Surveillance of People and Their Activities[J].IEEE Transaction on Pattern Analysis and Machine Intelligence,2000,22(8).
[38]钱晋.基于背景差法的视频车辆检测算法研究[D].上海:上海交通大学,2007.
[39]李扬.视频检测中高斯分布模型的应用[D].长春:吉林大学,2007.
[40]谭文明.复杂背景建模方法及其在运动目标检测中的应用[D].长沙:国防科学技术大学,2009.
[41] Stauffer C,WEL Crimson.Adaptive background mixture models for real-time tracking[C].Proc of the IEEE Conference on Computer Vision and Pattern Recognition,1999,2:246-252.
[42] Power P W,Schoonees J A.Understanding background mixture models for foregrounds segmentation[C].In: Proceedings of Image and Vision Computing. New Zealand: Auckland,2002,267-271.
[43] Nir Friedman,Stuart Russell.Image segmentation in video sequences:A probabilistic approach[C].Proc.of the 13th Conf on Uncertainty in Artificial Intelligence,1997,97:175-181.
[44] M.Piccardi.Background Subtraction Techniques:A Review[C].2004 IEEE International Conference on Systems,Man and Cybernetics,2004,3099-3104.
[45] C.Stauffer,W.Eric,L.Grimson.Learning Patterns of Activity Using Real-Time Tracking[J].IEEE Transactions on Pattern Analysis and Machine Intelligence, 2000, 8, (22):747-757.