基于视频分析的客流检测子系统的设计与实现
|
摘要
近年来随着我国城市化进程的加速,城市轨道交通得到快速发展,建设规模迅速扩大,交通路网运营客流量逐年日益增大,网络化运营也已成为我国城市轨道交通发展的必然趋势,同时网络化运营对信息共享和系统安全保障等提出新的更高要求。本文所涉及的客流状态实时检测与预警系统来源于国家科技支撑计划中的“城轨交通路网运营安全保障关键技术与系统研制”课题中的子课题的子任务,即通过获取、接入和集成包括路网基础设施、运营车辆状态、运输组织方案、客流分布等各个方面的数据信息,最终构建城轨交通路网运营数据中心。
本文的主要任务是以数字图像处理和机器学习理论为基础,对摄像头监控范围内的客流量进行统计,并最终实现客流状态实时监测与预警系统中的客流检测子系统,为构建城轨交通路网运营数据中心提供数据支持。本文的主要工作以目标检测为中心,围绕机器学习和视频分析两大模块进行设计与实现,机器学习模块主要完成了HOG特征提取和训练SVM分类器设计和实现。视频分析模块则首先论述HOG特征提取的快速算法,有效地提高了目标在线检测的速度。接着使用了金字塔多分辨率检测法对检测区域内不同尺寸的目标进行了检测。最后提出了多目标融合方案,并论述了多目标融合的具体实现流程。以上所有任务均由本人独立完成。
目前本系统正处于测试完善阶段,还未正式上线使用。通过测试显示,系统对监控范围内的目标检测率在正常光照下达到95%以上,当光照明显降低和增强时,会导致系统检测率出现不同程度下降。检测速度也达到了每秒2帧。客流状态实时检测与预警系统的研发为城轨运营管理部门提供各检测点客流分布元数据,为其制定有效的安全运营管理、应急救援和联动协调方案提供有力的技术支持。
With the acceleration of urbanization process in recent years, urban rail transit construction has expanded rapidly and the burden of the traffic network operators increase year by year. The networked operation has become the inevitable trend of developmenting urban rail transit, but meanwhile networked operation has a new and higher requirement for information sharing and system security. The real-time passenger state detection and early warning system this paper involved come from a subtopic named "urban rail transport network operational security and key technology study" which is from National Science and Technology Supported Program, that is, through acquiring, accessing and integrating the data from road network infrastructure, operating vehicle state, transport organizations scheme and passenger distribution to ultimately build the Urban Rail Network Operation Data Center.
The prime task of this paper is based on digital image processing and machine learning theory, to sum up the passenger flow within the monitored area. Ultimately, to build Passenger Flow Detection Subsystem. Targets detecting, which is the main work of this paper, centered on designing and implementing of two modules, machine learning and video analysis. The machine learning module completed HOG Feature Extraction and Training SVM classifier; Video analysis module first discusses a fast algorithm of HOG feature extraction, to improve the speed of the target detection. Then use the pyramid multi-resolution method to detect different sizes target within the detection area. Finally, introduce the multi-target fusion method and the detailed flow to realize. All the tasks mentioned above are completed by myself.
The system is in beta and perfect stage has not yet formally launched. Testing has shown that the system's detection rate has reached more than95%in normal light. When the light is significantly reduced and enhanced, will lead to the detection rate decreased to varying degrees. The Passenger status real-time detection and early warning system provide the passenger flow distribution data of each detection point for urban rail operations management, and provide strong technical support for its development of effective security operations management, emergency rescue and joint coordination scheme.
本文的主要任务是以数字图像处理和机器学习理论为基础,对摄像头监控范围内的客流量进行统计,并最终实现客流状态实时监测与预警系统中的客流检测子系统,为构建城轨交通路网运营数据中心提供数据支持。本文的主要工作以目标检测为中心,围绕机器学习和视频分析两大模块进行设计与实现,机器学习模块主要完成了HOG特征提取和训练SVM分类器设计和实现。视频分析模块则首先论述HOG特征提取的快速算法,有效地提高了目标在线检测的速度。接着使用了金字塔多分辨率检测法对检测区域内不同尺寸的目标进行了检测。最后提出了多目标融合方案,并论述了多目标融合的具体实现流程。以上所有任务均由本人独立完成。
目前本系统正处于测试完善阶段,还未正式上线使用。通过测试显示,系统对监控范围内的目标检测率在正常光照下达到95%以上,当光照明显降低和增强时,会导致系统检测率出现不同程度下降。检测速度也达到了每秒2帧。客流状态实时检测与预警系统的研发为城轨运营管理部门提供各检测点客流分布元数据,为其制定有效的安全运营管理、应急救援和联动协调方案提供有力的技术支持。
With the acceleration of urbanization process in recent years, urban rail transit construction has expanded rapidly and the burden of the traffic network operators increase year by year. The networked operation has become the inevitable trend of developmenting urban rail transit, but meanwhile networked operation has a new and higher requirement for information sharing and system security. The real-time passenger state detection and early warning system this paper involved come from a subtopic named "urban rail transport network operational security and key technology study" which is from National Science and Technology Supported Program, that is, through acquiring, accessing and integrating the data from road network infrastructure, operating vehicle state, transport organizations scheme and passenger distribution to ultimately build the Urban Rail Network Operation Data Center.
The prime task of this paper is based on digital image processing and machine learning theory, to sum up the passenger flow within the monitored area. Ultimately, to build Passenger Flow Detection Subsystem. Targets detecting, which is the main work of this paper, centered on designing and implementing of two modules, machine learning and video analysis. The machine learning module completed HOG Feature Extraction and Training SVM classifier; Video analysis module first discusses a fast algorithm of HOG feature extraction, to improve the speed of the target detection. Then use the pyramid multi-resolution method to detect different sizes target within the detection area. Finally, introduce the multi-target fusion method and the detailed flow to realize. All the tasks mentioned above are completed by myself.
The system is in beta and perfect stage has not yet formally launched. Testing has shown that the system's detection rate has reached more than95%in normal light. When the light is significantly reduced and enhanced, will lead to the detection rate decreased to varying degrees. The Passenger status real-time detection and early warning system provide the passenger flow distribution data of each detection point for urban rail operations management, and provide strong technical support for its development of effective security operations management, emergency rescue and joint coordination scheme.
引文
[1]孙章.城市轨道交通研究[J].北京.科学出版社.2012.2(2).45-46
[2]柏柯嘉.视频目标跟踪关键问题的研究[D].广州.华南理L大学.2009.
[3]周健.基于梯度特征和级联分类器的快速行人检测[D].天津：天津大学.2010.05
[4]Rafael C. Gonzalez and Richard E.Wood数字图像处理(阮秋奇,阮宇智等译)[M].北京：电子工业出版社.2010.1.70-74
[5]Dalal N and Triggs B. Histogram of oriented gradients for human detection[D].In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.2005:886-893
[6]Dalal N and Triggs B. Human Detection Using Oriented Histograms of Flow and Appearance[D]. In Proceeding of Europeon Conference on Computer Vision(2).2006:428-441
[7]Qiang Zhu, Shai Avidan,Mei-Chen Yeh and Kwang-Ting Cheng. Fast Human Detection Using a Cascade of Histograms of Oriented Gradients[D]. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition(6).2006
[8]Tom M. Mitchell机器学习(曾华军,张银奎等译)[M].北京：机械工业出版社.2003.1.23-27
[9]F. Suard, A. Rakotomamonjy, A. Bensrhair. Pedestrian Detection using Infrared Images and Histograms of Oriented Gradients[D]. In:Bertozzi M eds. Proceedings IEEE of Intelligent Vehicles Symposium, 2003.06. 662-667
[10]Sabzmeydani. P. and Mori. G. Detecting pedestrians by learning shapelet features[D]. In: Proceedings of IEEE Conference on Computer Vision and Pattern Recognition. USA: Minneapolis 2 007. 1-8
[11]王书舟.支撑向量机方法及其应用[D].哈尔滨：哈尔滨工业大学.2009.05.20-22
[12]克里斯特安尼.支撑向量机导论(李国正,王猛,曾华军泽)[M].北京：电子工业出版社.2004.2.35-37
[13]SVM八股简介:http://wenku.baidu.com/view/6add53c608a1284ac850430e.html
[14]Shigeo Abe. Surpport Vector Machines for Pattern Classification[M]. New York. Spring Science and Business Media,LLC,2010.09.21-103
[15]Christopher M.Bishop. Pattern Recognition and Machine Learning[M]. New York. Spring Science and Business Media,LLC,2006. 291-321
[16]陈光英,张千里,李星.特征选择和SVM训练模型的联合优化[J].清华大学学报.2004(1).9-12
[17]张铮,王艳平.数字图像处理与机器视觉Visual C++与Matlab实现[M].北京：人民邮电出版社.2010.04
[18]钟必能.复杂动态场景中运动目标检测与跟踪算法研究[D].哈尔滨：哈尔滨工业大学.2010.09
[19]汤义.智能交通系统中基于视频的行人检测和跟踪方法研究[D].2010.09
[20]谭恒良.基于融合全局与局部HOG特征的人脸识别方法[D].广州：中山大学.2011.04
[21]王文勇.基于视频分析的智能视频监控系统的研究与设计[D].成都：电子科技大学.2011.05
[22]黄亮.基于双目视觉的客流统计识别系统[D].上海：上海交通大学.2008.01
[23]车志富.基于支持向量机的行人检测[D].北京：北京交通大学.2010.06
[24]胡志.基于支持向量新颖检测[D].西安：西安电子科技大学.2010.01
[25]沈海超.嵌入式视频分析系统软什平台的设计与实现[D].太原：太原理工大学.2011.05
[26]王永志.视频场快速行人动态检测算法与实现[D].2011.04
[27]汤义.智能交通系统中基于视频的行人检测和跟踪方法研究[D].广州：华南理工大学.2010.09
[28]吴志刚.智能监控系统中目标检测算法的研究[D].武汉：武汉理工大学2009.05
[2]柏柯嘉.视频目标跟踪关键问题的研究[D].广州.华南理L大学.2009.
[3]周健.基于梯度特征和级联分类器的快速行人检测[D].天津：天津大学.2010.05
[4]Rafael C. Gonzalez and Richard E.Wood数字图像处理(阮秋奇,阮宇智等译)[M].北京：电子工业出版社.2010.1.70-74
[5]Dalal N and Triggs B. Histogram of oriented gradients for human detection[D].In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.2005:886-893
[6]Dalal N and Triggs B. Human Detection Using Oriented Histograms of Flow and Appearance[D]. In Proceeding of Europeon Conference on Computer Vision(2).2006:428-441
[7]Qiang Zhu, Shai Avidan,Mei-Chen Yeh and Kwang-Ting Cheng. Fast Human Detection Using a Cascade of Histograms of Oriented Gradients[D]. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition(6).2006
[8]Tom M. Mitchell机器学习(曾华军,张银奎等译)[M].北京：机械工业出版社.2003.1.23-27
[9]F. Suard, A. Rakotomamonjy, A. Bensrhair. Pedestrian Detection using Infrared Images and Histograms of Oriented Gradients[D]. In:Bertozzi M eds. Proceedings IEEE of Intelligent Vehicles Symposium, 2003.06. 662-667
[10]Sabzmeydani. P. and Mori. G. Detecting pedestrians by learning shapelet features[D]. In: Proceedings of IEEE Conference on Computer Vision and Pattern Recognition. USA: Minneapolis 2 007. 1-8
[11]王书舟.支撑向量机方法及其应用[D].哈尔滨：哈尔滨工业大学.2009.05.20-22
[12]克里斯特安尼.支撑向量机导论(李国正,王猛,曾华军泽)[M].北京：电子工业出版社.2004.2.35-37
[13]SVM八股简介:http://wenku.baidu.com/view/6add53c608a1284ac850430e.html
[14]Shigeo Abe. Surpport Vector Machines for Pattern Classification[M]. New York. Spring Science and Business Media,LLC,2010.09.21-103
[15]Christopher M.Bishop. Pattern Recognition and Machine Learning[M]. New York. Spring Science and Business Media,LLC,2006. 291-321
[16]陈光英,张千里,李星.特征选择和SVM训练模型的联合优化[J].清华大学学报.2004(1).9-12
[17]张铮,王艳平.数字图像处理与机器视觉Visual C++与Matlab实现[M].北京：人民邮电出版社.2010.04
[18]钟必能.复杂动态场景中运动目标检测与跟踪算法研究[D].哈尔滨：哈尔滨工业大学.2010.09
[19]汤义.智能交通系统中基于视频的行人检测和跟踪方法研究[D].2010.09
[20]谭恒良.基于融合全局与局部HOG特征的人脸识别方法[D].广州：中山大学.2011.04
[21]王文勇.基于视频分析的智能视频监控系统的研究与设计[D].成都：电子科技大学.2011.05
[22]黄亮.基于双目视觉的客流统计识别系统[D].上海：上海交通大学.2008.01
[23]车志富.基于支持向量机的行人检测[D].北京：北京交通大学.2010.06
[24]胡志.基于支持向量新颖检测[D].西安：西安电子科技大学.2010.01
[25]沈海超.嵌入式视频分析系统软什平台的设计与实现[D].太原：太原理工大学.2011.05
[26]王永志.视频场快速行人动态检测算法与实现[D].2011.04
[27]汤义.智能交通系统中基于视频的行人检测和跟踪方法研究[D].广州：华南理工大学.2010.09
[28]吴志刚.智能监控系统中目标检测算法的研究[D].武汉：武汉理工大学2009.05