智能监控系统中的背景建模算法研究
|
摘要
近年来,随着电子技术、计算机技术、网络通信技术、图像和视频处理技术的不断进步,基于图像和视频处理技术的智能视频监控系统取得了很大的发展,并且在公共安全方面发挥着越来越重要的作用。与此同时,对智能监控系统的研究也日益成为备受关注的前沿课题,与智能监控技术紧密相关的背景建模等目标检测算法也成为相关领域内研究的热点话题。
论文对常用的背景建模算法进行了深入分析与建模仿真,自行设计搭建智能监控系统硬件平台,并在该平台上运行背景建模算法实现目标检测,完成的主要工作内容如下：
(1)搭建基于DSP+FPGA架构的智能监控系统硬件平台。其中DSP负责运行背景建模算法,FPGA负责实现外部接口电路。外部接口电路主要包括：视频采集与输出接口、HDMI接口、LVDS接口等。该平台可实现多通道多格式视频图像采集、输出与视频图像处理等功能,还具有可组网功能,能构成分布式智能监控系统。
(2)背景建模算法仿真及分析。采集室外视频图像对目前常用的两类背景建模算法进行仿真实验,主要包括：基于统计模型的背景建模算法与直接计算图像像素值的背景建模算法。仿真结果显示处理背景扰动越强则算法实时性能越差,算法实时性能好的建模算法往往建模准确性不佳。
(3)背景建模算法移植及硬件实现。在分析比较了各种算法后,综合考虑硬件平台内存容量、算法缓存需求、实时性能等因素,选取帧间差分法作为最终算法,并将其移植到DSP+FPGA硬件平台上进行验证。最终实验结果显示背景建模效果稳定,目标检测与追踪效果良好。
In recent years, electronic, computer, network communication, image processing and other related fields have made a lot of development. With the development, intelligent video surveillance system which based on image and video processing has made a lot of progress, and the system is playing a more importantly role in the public safety area. At the same time, the technology which closely related to the intelligent surveillance technology has become a hot research topic, such as target detection and background modeling.
In order to realize intelligent target detection, this paper carried out modeling and simulation to the popular background modeling algorithm; designed the hardware platform of the intelligent surveillance system; and run the background modeling algorithm in the hardware platform at last. Completion of the main tasks was shown as follows:
(1) Build intelligent surveillance system hardware platform which based on DSP+FPGA. In the hardware platform, DSP was responsible for running background modeling algorithm; FPGA was responsible for the external interface circuit. The external interface circuit including:video capture and output, HDMI, LVDS and so on. The platform can achieve multi-channel, multi-format video image capture, output and video image processing, and it also can be networking features, which could form a distributed intelligent surveillance system.
(2) Simulation and analysis of background modeling algorithm. In this paper, background modeling algorithm was divided into two categories:one is based on statistical models and another is calculating pixel value of background directly. The simulation results shows that if the background modeling algorithm could handle the background disturbance better, the real-time performance will be worse, and if the real-time performance better, modeling accuracy will be poorer.
(3) Background modeling algorithm transplant and the hardware implementation. After had analyzed various algorithms, and considering the memory capacity of the hardware platform, the algorithm cache demand, real-time performance and other factors, this paper select the frame difference method as the final algorithm to transplant to the DSP+FPGA hardware platform for validation. The final experimental results show that the effect of background modeling is stable, and the target detection and tracking also have good effect.
论文对常用的背景建模算法进行了深入分析与建模仿真,自行设计搭建智能监控系统硬件平台,并在该平台上运行背景建模算法实现目标检测,完成的主要工作内容如下：
(1)搭建基于DSP+FPGA架构的智能监控系统硬件平台。其中DSP负责运行背景建模算法,FPGA负责实现外部接口电路。外部接口电路主要包括：视频采集与输出接口、HDMI接口、LVDS接口等。该平台可实现多通道多格式视频图像采集、输出与视频图像处理等功能,还具有可组网功能,能构成分布式智能监控系统。
(2)背景建模算法仿真及分析。采集室外视频图像对目前常用的两类背景建模算法进行仿真实验,主要包括：基于统计模型的背景建模算法与直接计算图像像素值的背景建模算法。仿真结果显示处理背景扰动越强则算法实时性能越差,算法实时性能好的建模算法往往建模准确性不佳。
(3)背景建模算法移植及硬件实现。在分析比较了各种算法后,综合考虑硬件平台内存容量、算法缓存需求、实时性能等因素,选取帧间差分法作为最终算法,并将其移植到DSP+FPGA硬件平台上进行验证。最终实验结果显示背景建模效果稳定,目标检测与追踪效果良好。
In recent years, electronic, computer, network communication, image processing and other related fields have made a lot of development. With the development, intelligent video surveillance system which based on image and video processing has made a lot of progress, and the system is playing a more importantly role in the public safety area. At the same time, the technology which closely related to the intelligent surveillance technology has become a hot research topic, such as target detection and background modeling.
In order to realize intelligent target detection, this paper carried out modeling and simulation to the popular background modeling algorithm; designed the hardware platform of the intelligent surveillance system; and run the background modeling algorithm in the hardware platform at last. Completion of the main tasks was shown as follows:
(1) Build intelligent surveillance system hardware platform which based on DSP+FPGA. In the hardware platform, DSP was responsible for running background modeling algorithm; FPGA was responsible for the external interface circuit. The external interface circuit including:video capture and output, HDMI, LVDS and so on. The platform can achieve multi-channel, multi-format video image capture, output and video image processing, and it also can be networking features, which could form a distributed intelligent surveillance system.
(2) Simulation and analysis of background modeling algorithm. In this paper, background modeling algorithm was divided into two categories:one is based on statistical models and another is calculating pixel value of background directly. The simulation results shows that if the background modeling algorithm could handle the background disturbance better, the real-time performance will be worse, and if the real-time performance better, modeling accuracy will be poorer.
(3) Background modeling algorithm transplant and the hardware implementation. After had analyzed various algorithms, and considering the memory capacity of the hardware platform, the algorithm cache demand, real-time performance and other factors, this paper select the frame difference method as the final algorithm to transplant to the DSP+FPGA hardware platform for validation. The final experimental results show that the effect of background modeling is stable, and the target detection and tracking also have good effect.
引文
[1]潘国辉.智能网络视频监控技术详解与实践[M].北京：清华大学出版社.2010.
[2]汪光华.智能安防—视频监控全面解析与实例分析[M].北京：机械工业出版社.2012.
[3]庞伟.中国监控监视技术产品市场分析[J].中国公共全,2009,28(6)：34-44.
[4]汪光华.视频监控系统应用[M].北京：中国政法大学出版社.2009.
[5]方帅.计算机智能视频监控系统的关键技术研究[D].沈阳：东北大学.2005.
[6]张伯虎,杨金柱,陈建莉等.智能视频监控系统的分析与应用[J].安防科技.2006,11：9-11.
[7]Collins R et al.A system for video surveillance and monitoring:VSAM final report, Camegie Mellon University, Technical Report:CMU-RI-TR-00-12,2000.
[8]徐珊珊.基于运动目标分析的智能监控系统[D].北京：北京邮电大学.2010.
[9]余静,游志胜.自动目标识别与跟踪技术研究综述[J].计算机应用研究.2005.22(1)：12-15.
[10]王新余,张桂林.基于光流的运动动目标实时监测方法研究[J]计算机工程与应用.2004.1：35-28.
[11]刘亚丽.背景建模技术的研究与实现[D].北京：北方工业大学.2010.
[12]朱明旱,罗大庸,曹倩霞.帧间差分与背景差分相融合的运动目标检测算法[J].自动化测试.2005.13(3)：215-221.
[13]李秀峰,苏兰海.均值滤波算法及应用研究[J].自动化应用.2008.23(7)：113-116.
[14]Kentaro Toyama, John Krumm, Barry Brumitt. Wallflower:Principles and Practice of Background Maintenance[J]. International Conference on Computer Vision,1999,23(6):216-228.
[15]Ismail Haritaoglu, David Harwood, Larry S. Davis. W4:Real-Time Surveillance of People and Their Activities, Transaction on Pattern Analysis and Machine Intelligence,2000.22(8):809-829.
[16]甘新胜.基于码书的运动目标检测方法[J].中国图像图形学报,2008.13(2)：365-372.
[17]Ahmed Elgammal, Ramani Duraiswami. Background and Foreground Modeling Using Nonparametric Kernel Density Estimation for Visual Surveillance. Proceedings of the IEEE,2002.90(7):1151-1164.
[18]张志付,付宇卓.一种基于背景减法的运动检测算法评价方法[J].信息技术.2008.12(3)：52-56.
[19]Zoran Zivkovic. Improved Adaptive Gaussian Mixture Model for Background Subtraction. InProc.ICPR,2004.
[20]查成东,王长松.基于自适应背景模型的运动目标检测[J].光电工程,2008.35(1)：26-30.
[21]何明一,卫保国.数字图像处理[M].北京：科学出版社.2008.
[22]贾永红.计算机图像处理与分析[M].北京：清华大学出版社.2001.
[23]Ismail Haritaoglu, David Harwood, Larry S. W4:Who, When, Where, What:A Real Time System for Detcting and Tracking People. Third Face and Gesture Recognition Conf,1998:222-227.
[24]贺彪.动态背景下基于码书模型的运动目标检测[D].安徽：安徽大学.2011.
[25]W.E.L Grimson, C. Stauffer, R. Romano, and L. Lee. Using adaptive tracking to classify and monitor activities in a site. In Computer Vision and Pattern Recognition,1998.22-29,
[26]Ahmed Elgammal, David Harwood, Larry S. Non-parametric Model for Background Subtraction. Eurpopean Conference on Computer Vision.2000.
[27]顾凤嫱.基于Mean shift的视频跟踪算法研究[D].陕西：西安电子科技大学.2009.
[28]代李学,李国辉,涂丹等.监控视运动目标检测减背景技术的研究现状和展望[J].中国图像图形学报,2006.211(7)：919-927.
[2]汪光华.智能安防—视频监控全面解析与实例分析[M].北京：机械工业出版社.2012.
[3]庞伟.中国监控监视技术产品市场分析[J].中国公共全,2009,28(6)：34-44.
[4]汪光华.视频监控系统应用[M].北京：中国政法大学出版社.2009.
[5]方帅.计算机智能视频监控系统的关键技术研究[D].沈阳：东北大学.2005.
[6]张伯虎,杨金柱,陈建莉等.智能视频监控系统的分析与应用[J].安防科技.2006,11：9-11.
[7]Collins R et al.A system for video surveillance and monitoring:VSAM final report, Camegie Mellon University, Technical Report:CMU-RI-TR-00-12,2000.
[8]徐珊珊.基于运动目标分析的智能监控系统[D].北京：北京邮电大学.2010.
[9]余静,游志胜.自动目标识别与跟踪技术研究综述[J].计算机应用研究.2005.22(1)：12-15.
[10]王新余,张桂林.基于光流的运动动目标实时监测方法研究[J]计算机工程与应用.2004.1：35-28.
[11]刘亚丽.背景建模技术的研究与实现[D].北京：北方工业大学.2010.
[12]朱明旱,罗大庸,曹倩霞.帧间差分与背景差分相融合的运动目标检测算法[J].自动化测试.2005.13(3)：215-221.
[13]李秀峰,苏兰海.均值滤波算法及应用研究[J].自动化应用.2008.23(7)：113-116.
[14]Kentaro Toyama, John Krumm, Barry Brumitt. Wallflower:Principles and Practice of Background Maintenance[J]. International Conference on Computer Vision,1999,23(6):216-228.
[15]Ismail Haritaoglu, David Harwood, Larry S. Davis. W4:Real-Time Surveillance of People and Their Activities, Transaction on Pattern Analysis and Machine Intelligence,2000.22(8):809-829.
[16]甘新胜.基于码书的运动目标检测方法[J].中国图像图形学报,2008.13(2)：365-372.
[17]Ahmed Elgammal, Ramani Duraiswami. Background and Foreground Modeling Using Nonparametric Kernel Density Estimation for Visual Surveillance. Proceedings of the IEEE,2002.90(7):1151-1164.
[18]张志付,付宇卓.一种基于背景减法的运动检测算法评价方法[J].信息技术.2008.12(3)：52-56.
[19]Zoran Zivkovic. Improved Adaptive Gaussian Mixture Model for Background Subtraction. InProc.ICPR,2004.
[20]查成东,王长松.基于自适应背景模型的运动目标检测[J].光电工程,2008.35(1)：26-30.
[21]何明一,卫保国.数字图像处理[M].北京：科学出版社.2008.
[22]贾永红.计算机图像处理与分析[M].北京：清华大学出版社.2001.
[23]Ismail Haritaoglu, David Harwood, Larry S. W4:Who, When, Where, What:A Real Time System for Detcting and Tracking People. Third Face and Gesture Recognition Conf,1998:222-227.
[24]贺彪.动态背景下基于码书模型的运动目标检测[D].安徽：安徽大学.2011.
[25]W.E.L Grimson, C. Stauffer, R. Romano, and L. Lee. Using adaptive tracking to classify and monitor activities in a site. In Computer Vision and Pattern Recognition,1998.22-29,
[26]Ahmed Elgammal, David Harwood, Larry S. Non-parametric Model for Background Subtraction. Eurpopean Conference on Computer Vision.2000.
[27]顾凤嫱.基于Mean shift的视频跟踪算法研究[D].陕西：西安电子科技大学.2009.
[28]代李学,李国辉,涂丹等.监控视运动目标检测减背景技术的研究现状和展望[J].中国图像图形学报,2006.211(7)：919-927.