基于视频虚拟线圈的交通流参数检测
|
摘要
实时交通流数据的采集，包括车流量统计、车道平均车速、车型分类、车道占有率等，在智能交通系统中起着重要的作用。交通流检测器有电磁感应线圈、超声波检测器、微波检测器和红外光标等多种方式，而基于图像处理的视频检测方式近年发展很快，因为它具有大区域检测、设置灵活等优越性，已成为智能交通系统的一个研究热点。
通常的视频检测手段强调二维图像的处理算法，包括车辆的识别、分割和跟踪等。本文致力于研究一种灵活并且可靠的视频交通流检测系统，在图像中使用智能传感器的概念，完成自动快速的数据运算。其核心思想是通过视频虚拟线圈将二维的图像信号转化成一维的数字时间信号。虚拟线圈的作用类似于电磁感应线圈，每个车道可以设置一到两个虚拟线圈来检测交通流参数，该方法的特点是避开了在二维图像空间中进行复杂的车辆特征提取与跟踪。
虚拟线圈可以在视频图像中自由设置，并且位置和大小可以调整。系统只处理虚拟线圈内的图像，由此降低了运算时间。各个虚拟线圈的输出信号主要来源于帧间差分，当帧间差分的结果小于判断阈值时，系统会自动调用减背景图像处理方法来产生虚拟线圈信号。通过对交通流实际样本的自学习，系统能够找到合适的放大倍数来调整线圈输出信号，从而抑制噪声，使得虚拟线圈的输出信号更真实的反映车辆的实际通过情况。
每个虚拟线圈都拥有自己的数据缓冲池而不会造成多线圈间的数据访问混乱，数据缓冲池中的数据是逐帧更新的。通过监视各个数据缓冲池中的信号，系统能够检测到车辆的出现并开启跟踪进程。当同一个车道上前后两个线圈同时检测到车辆的运动时，系统从两个线圈的数据缓冲池中找到相应索引位置的数据进行相关匹配求得车辆的行驶速度。为了精确的在一维时间信号中跟踪车辆，系统引入了两个重要的参数（最大容错度和最小可靠性）以及跟踪进程中的奖惩策略。最后，各车道的流量统计、各车辆的速度测量以及车型分类通过监视、跟踪虚拟线圈的输出信号求得。
视频虚拟线圈检测交通流参数的实时处理原型系统已经在计算机上实现，实验结果表明该系统具有较高的车流量和车速检测精度。
The collection of real-time traffic data, such as traffic load, average travel speed, vehicle classification, and lane occupancy, plays a critical role in the advanced traffic management system and traveler information systems. Technological innovations have given rise to different types of traffic detectors. Conventional detectors, for example, inductive loop, detectors using ultrasonic, microwave, or infrared, have been put into use for several decades. Meanwhile, a promising approach, video-based measurement system, has developed quickly. Since it has many advantages, for instance, wider-area detection and superior flexibility, many researches have been done in this area.
Previous methods are mainly based on image processing algorithms, especially on vehicle extraction and vehicle tracking. This paper is focused on developing a flexible and reliable system to detect the traffic parameters through image sequences. One of the system's ultimate goals is on intelligent image sensors and automated fast data processing. The key idea of the system is converting the two-dimensional image data to one-dimensional digital temporal signals by virtual-loops. Virtual-loop's function is similar with the inductive loop sensor. Each lane can have one or two virtual-loops to detect its traffic parameters. This method avoids the complicated vehicle identification, extraction and tracking.
Virtual-loop can be freely set in the video image with its position and size adjustable. Only the image data within the virtual-loops are processed, so the time cost of calculation is reduced. Each virtual-loop's output signals mainly derive from the pixel difference between consecutive image frames within the virtual-loop area. When the result of consecutive frame difference is smaller than the threshold, current frame subtracts the background to produce the virtual-loop's signals. Through studying the traffic sample, the system itself can find the appropriate parameter to adjust the virtual-loop's signals, which restrains the noise's effect. Thus the virtual-loop's signals can reflect the passing vehicles accurately.
Each virtual-loop's output signals are recorded in its special data buffer, which is updated when processing every frame. Through observing the digital string of each
data buffer, system can detect the vehicle's appearance and track it. When two virtual-loops in the same lane all detect one vehicle's motion, system retrieve the necessary data from these two virtual-loop's data buffer to calculate the vehicle's velocity by signal correlation. In order to tracking the vehicle more inerrably, system uses two important parameters (maximum error tolerance and minimum reliability) and an effective algorithm (punishment and reward algorithm). In the end, the parameters of each lane's traffic load, vehicle velocity and vehicle classification are acquired through the surveillance of the virtual-loops output signals.
The real-time prototype system is implemented on a computer. Experimental results illustrate the detection accuracy.
通常的视频检测手段强调二维图像的处理算法，包括车辆的识别、分割和跟踪等。本文致力于研究一种灵活并且可靠的视频交通流检测系统，在图像中使用智能传感器的概念，完成自动快速的数据运算。其核心思想是通过视频虚拟线圈将二维的图像信号转化成一维的数字时间信号。虚拟线圈的作用类似于电磁感应线圈，每个车道可以设置一到两个虚拟线圈来检测交通流参数，该方法的特点是避开了在二维图像空间中进行复杂的车辆特征提取与跟踪。
虚拟线圈可以在视频图像中自由设置，并且位置和大小可以调整。系统只处理虚拟线圈内的图像，由此降低了运算时间。各个虚拟线圈的输出信号主要来源于帧间差分，当帧间差分的结果小于判断阈值时，系统会自动调用减背景图像处理方法来产生虚拟线圈信号。通过对交通流实际样本的自学习，系统能够找到合适的放大倍数来调整线圈输出信号，从而抑制噪声，使得虚拟线圈的输出信号更真实的反映车辆的实际通过情况。
每个虚拟线圈都拥有自己的数据缓冲池而不会造成多线圈间的数据访问混乱，数据缓冲池中的数据是逐帧更新的。通过监视各个数据缓冲池中的信号，系统能够检测到车辆的出现并开启跟踪进程。当同一个车道上前后两个线圈同时检测到车辆的运动时，系统从两个线圈的数据缓冲池中找到相应索引位置的数据进行相关匹配求得车辆的行驶速度。为了精确的在一维时间信号中跟踪车辆，系统引入了两个重要的参数（最大容错度和最小可靠性）以及跟踪进程中的奖惩策略。最后，各车道的流量统计、各车辆的速度测量以及车型分类通过监视、跟踪虚拟线圈的输出信号求得。
视频虚拟线圈检测交通流参数的实时处理原型系统已经在计算机上实现，实验结果表明该系统具有较高的车流量和车速检测精度。
The collection of real-time traffic data, such as traffic load, average travel speed, vehicle classification, and lane occupancy, plays a critical role in the advanced traffic management system and traveler information systems. Technological innovations have given rise to different types of traffic detectors. Conventional detectors, for example, inductive loop, detectors using ultrasonic, microwave, or infrared, have been put into use for several decades. Meanwhile, a promising approach, video-based measurement system, has developed quickly. Since it has many advantages, for instance, wider-area detection and superior flexibility, many researches have been done in this area.
Previous methods are mainly based on image processing algorithms, especially on vehicle extraction and vehicle tracking. This paper is focused on developing a flexible and reliable system to detect the traffic parameters through image sequences. One of the system's ultimate goals is on intelligent image sensors and automated fast data processing. The key idea of the system is converting the two-dimensional image data to one-dimensional digital temporal signals by virtual-loops. Virtual-loop's function is similar with the inductive loop sensor. Each lane can have one or two virtual-loops to detect its traffic parameters. This method avoids the complicated vehicle identification, extraction and tracking.
Virtual-loop can be freely set in the video image with its position and size adjustable. Only the image data within the virtual-loops are processed, so the time cost of calculation is reduced. Each virtual-loop's output signals mainly derive from the pixel difference between consecutive image frames within the virtual-loop area. When the result of consecutive frame difference is smaller than the threshold, current frame subtracts the background to produce the virtual-loop's signals. Through studying the traffic sample, the system itself can find the appropriate parameter to adjust the virtual-loop's signals, which restrains the noise's effect. Thus the virtual-loop's signals can reflect the passing vehicles accurately.
Each virtual-loop's output signals are recorded in its special data buffer, which is updated when processing every frame. Through observing the digital string of each
data buffer, system can detect the vehicle's appearance and track it. When two virtual-loops in the same lane all detect one vehicle's motion, system retrieve the necessary data from these two virtual-loop's data buffer to calculate the vehicle's velocity by signal correlation. In order to tracking the vehicle more inerrably, system uses two important parameters (maximum error tolerance and minimum reliability) and an effective algorithm (punishment and reward algorithm). In the end, the parameters of each lane's traffic load, vehicle velocity and vehicle classification are acquired through the surveillance of the virtual-loops output signals.
The real-time prototype system is implemented on a computer. Experimental results illustrate the detection accuracy.
引文
[1] 史其信, 陆化普. 中国ITS发展战略构想. 公路交通科技. 1998. 15(3):13~16
[2] 艾贺申, 高海龙. 美国智能运输系统(ITS)研究进展. 河南交通科技. 1999, 19(4): 41~43
[3] 尚刚, 陈宝. 智能交通系统(ITS)在日本的发展综述. 华东公路. 1999, 118(3): 62~65
[4] 岳云. 智能交通系统的发展近况. 科技与经济. 1998. No.1: 17~18
[5] Stephen J.Bahler. Field Test of non-intrusive traffic Detection technologies. Proceedings of the 5th World Congress on Intelligent Transportation Systems. Seoul, Korea, 1998
[6] Koji Aikawa. Introduction of Intelligent Vehicle Detection Sensors. Proceedings of the 5th World Congress on Intelligent Transportation Systems. Seoul, Korea, 1998
[7] 阙大顺. 车辆检测器及其进展研究. 交通与计算机. 1999, 17(2): 38~41.
[8] 王炜, 过秀成, 等. 交通工程学. 南京: 东南大学出版社, 2000
[9] 马骏. 高速公路行车安全距离的分析与研究. 西安公路交通大学学报. 1998, 18(4): 90~94
[10] 项乔君, 王炜, 李伟. 汽车最小行车时距计算模型的研究. 东南大学学报. 1998, 28(3): 79~82
[11] M.Y. Siyal, M.Fathy. A neural -vision based approach to measure traffic queue parameters in real-time. Pattern Recognition Letters. 20(1999): 761~770
[12] Charles Harlow, Shiquan Peng. Automatic vehicle calssfication system with range sensors. Transportation Research Part C. 9(2001): 231~247
[13] Chiaki Aoyama, Ken'ichi Mineta, Yasushi Okada. High-resolution range sensor system with stereo camera. JSAE(Society of Automotive Engineers of Japan) Review. 21(2000): 365~371
[14] G.D.Sullivan, K.D.baker, A.D.Worrall, et al. Model-based vehicle detection and classification using orthographic approximations. Image and Vision Computing. 15(1997): 649~654
[15] 刘国衡.图像序列的目标检测.计算机工程及应用.1997.8: 26~29
[16] 王栓.基于差分图像的多运动目标的检测与跟踪.中国图像图形学报. 1999.4(6): 470~475
[17] 尚裴,郭志芬.高速公路汽车目标检测.北京理工大学学报.1997,17(6): 759~754
[18] Yang-Ju Kim．Improvement of Background Update Method for Image Detector．5th World Congress on Intelligent Transport Systems
[19] 王俊杰,黄心汉.一种对图像进行快速二值化处理的方法.电子技术应用. 1998, No.10: 16~17
[20] 安玮.奇异点多门限图像分割算法.计算机工程与科学.1998,20(3):50~55
[21] 徐立亚.图像目标区域定位模糊法实现.红外与毫米波学报.1998，17(3):209~214
[22] 张玲华.对有背景的图像实现运动目标的边界检测.南京邮电学院学报. 1997,17(3): 77~79
[23] 邓英士.一种快速图像处理相关匹配方法.光电工程.1997,24(4):6~10
[24] Margrit Betke, Esin Haritaoglu, Larry S.Davis．Multiple vehicle detection and tracking．Proceedings of SPIE - The International Society for Optical Engineering Vol.2962 1997:104~110
[25] Benjamin Coifman, David Beymer, Philip McLauchlan, et al. Areal-time computer vision system for vehicle tracking and traffic surveillance. Transportation Research Part C 6 (1998):271~288
[26] Zhigang zhu, Guangyou Xu, Bo Yang, et al. A real-time vision system fro automatic traffic monitoring. Image and Vision Computing 18(2000): 781~794.
[27] Michael Haag, hans-Hellmut nagel. Tracking of complex driving manoeuvres in traffic image sequence. Image and Vision Computing 16(1998):517~527
[28] 刘君华.智能传感器系统.西安: 西安电子科技大学出版社.1999
[2] 艾贺申, 高海龙. 美国智能运输系统(ITS)研究进展. 河南交通科技. 1999, 19(4): 41~43
[3] 尚刚, 陈宝. 智能交通系统(ITS)在日本的发展综述. 华东公路. 1999, 118(3): 62~65
[4] 岳云. 智能交通系统的发展近况. 科技与经济. 1998. No.1: 17~18
[5] Stephen J.Bahler. Field Test of non-intrusive traffic Detection technologies. Proceedings of the 5th World Congress on Intelligent Transportation Systems. Seoul, Korea, 1998
[6] Koji Aikawa. Introduction of Intelligent Vehicle Detection Sensors. Proceedings of the 5th World Congress on Intelligent Transportation Systems. Seoul, Korea, 1998
[7] 阙大顺. 车辆检测器及其进展研究. 交通与计算机. 1999, 17(2): 38~41.
[8] 王炜, 过秀成, 等. 交通工程学. 南京: 东南大学出版社, 2000
[9] 马骏. 高速公路行车安全距离的分析与研究. 西安公路交通大学学报. 1998, 18(4): 90~94
[10] 项乔君, 王炜, 李伟. 汽车最小行车时距计算模型的研究. 东南大学学报. 1998, 28(3): 79~82
[11] M.Y. Siyal, M.Fathy. A neural -vision based approach to measure traffic queue parameters in real-time. Pattern Recognition Letters. 20(1999): 761~770
[12] Charles Harlow, Shiquan Peng. Automatic vehicle calssfication system with range sensors. Transportation Research Part C. 9(2001): 231~247
[13] Chiaki Aoyama, Ken'ichi Mineta, Yasushi Okada. High-resolution range sensor system with stereo camera. JSAE(Society of Automotive Engineers of Japan) Review. 21(2000): 365~371
[14] G.D.Sullivan, K.D.baker, A.D.Worrall, et al. Model-based vehicle detection and classification using orthographic approximations. Image and Vision Computing. 15(1997): 649~654
[15] 刘国衡.图像序列的目标检测.计算机工程及应用.1997.8: 26~29
[16] 王栓.基于差分图像的多运动目标的检测与跟踪.中国图像图形学报. 1999.4(6): 470~475
[17] 尚裴,郭志芬.高速公路汽车目标检测.北京理工大学学报.1997,17(6): 759~754
[18] Yang-Ju Kim．Improvement of Background Update Method for Image Detector．5th World Congress on Intelligent Transport Systems
[19] 王俊杰,黄心汉.一种对图像进行快速二值化处理的方法.电子技术应用. 1998, No.10: 16~17
[20] 安玮.奇异点多门限图像分割算法.计算机工程与科学.1998,20(3):50~55
[21] 徐立亚.图像目标区域定位模糊法实现.红外与毫米波学报.1998，17(3):209~214
[22] 张玲华.对有背景的图像实现运动目标的边界检测.南京邮电学院学报. 1997,17(3): 77~79
[23] 邓英士.一种快速图像处理相关匹配方法.光电工程.1997,24(4):6~10
[24] Margrit Betke, Esin Haritaoglu, Larry S.Davis．Multiple vehicle detection and tracking．Proceedings of SPIE - The International Society for Optical Engineering Vol.2962 1997:104~110
[25] Benjamin Coifman, David Beymer, Philip McLauchlan, et al. Areal-time computer vision system for vehicle tracking and traffic surveillance. Transportation Research Part C 6 (1998):271~288
[26] Zhigang zhu, Guangyou Xu, Bo Yang, et al. A real-time vision system fro automatic traffic monitoring. Image and Vision Computing 18(2000): 781~794.
[27] Michael Haag, hans-Hellmut nagel. Tracking of complex driving manoeuvres in traffic image sequence. Image and Vision Computing 16(1998):517~527
[28] 刘君华.智能传感器系统.西安: 西安电子科技大学出版社.1999