视频监控中运动目标检测与跟踪关键技术研究
|
摘要
视频监控中运动目标的检测与跟踪是计算机视觉和图像编码领域的重要研究项目之一,在军事、医学和科研等领域都有广泛的应用。运动目标检测与跟踪算法的设计直接影响跟踪效果的准确性和稳定性。本文主要研究视频监控中运动目标的检测与跟踪的关键问题。
在运动目标检测方面,首先对目前流行的帧间差法、背景差法、光流场法进行了实验分析和比较,指出其优缺点及适用范围:对于静态背景的获取、估计与更新进行了实验研究,引入了分块处理的思想,提出了分块背景估计算法,该方法增强了运动目标检测随环境变化的鲁棒性;提出一种基于码书的运动目标检测方法。该方法用矢量量化/聚类技术构建背景模型,利用当前帧和背景帧之间的亮度偏差和色度偏差来检测运动目标。实验表明,该方法具有很好的检测效果。
在运动目标跟踪方面,对常用的视频运动目标跟踪方法进行了分析比较,提出一种利用差分法和特征匹配进行目标跟踪的方法。该方法将差分法得到的目标用矩形框框起来,对矩形框内的目标求质心,然后利用质心结合形状特征进行匹配,实验表明,该方法可以较好的适应目标形状有一定变化的情况,简单易行,在情况不复杂的情况下可以较好地跟踪目标。最后,本文对Mean Shift目标跟踪方法从实现原理、匹配准则和搜索算法等几方面对进行了分析。
Motion object detection and tracking in serial images is the main research field in Video Surveillance, which has been widely applied in military, medicine and scientific research etc. The accuracy and stability of tracking effect depend on the design of algorithms to a great extent. The key technologies is studied in this paper about how to detect and track moving object in Video Surveillance.
On the research of the motion detection, firstly three main algorithms of motion detection and its analyzed advantages and drawbacks are researched. Secondly, background estimation based on block is proposed to solve the problem of Obtaining and estimating and updating background model which can enhance the robust of moving object detection. A background modeling and subtraction method by codebook construction is proposed. The CB algorithm adopts a quantization and clustering technique to construct a background model and subtract the current image from the background model to detect moving objects by color distortion and bright distortion. Evaluation shows that this algorithm is effective.
On the research of object tracking, firstly some common used algorithms of are researched. Then a new method of moving object tracking based on the result of image subtraction and object features matching is presented. The obtained object is framed and its center of gravity is calculated. In this together with its shape feather, its moving path can be obtained. Evaluation shows that this algorithm has better adoption in case of a little shape change and is effective in simple cases. At last the Mean Shift Algorithm is analyzed from the aspects of realization principle and match criterion and search algorithm.
在运动目标检测方面,首先对目前流行的帧间差法、背景差法、光流场法进行了实验分析和比较,指出其优缺点及适用范围:对于静态背景的获取、估计与更新进行了实验研究,引入了分块处理的思想,提出了分块背景估计算法,该方法增强了运动目标检测随环境变化的鲁棒性;提出一种基于码书的运动目标检测方法。该方法用矢量量化/聚类技术构建背景模型,利用当前帧和背景帧之间的亮度偏差和色度偏差来检测运动目标。实验表明,该方法具有很好的检测效果。
在运动目标跟踪方面,对常用的视频运动目标跟踪方法进行了分析比较,提出一种利用差分法和特征匹配进行目标跟踪的方法。该方法将差分法得到的目标用矩形框框起来,对矩形框内的目标求质心,然后利用质心结合形状特征进行匹配,实验表明,该方法可以较好的适应目标形状有一定变化的情况,简单易行,在情况不复杂的情况下可以较好地跟踪目标。最后,本文对Mean Shift目标跟踪方法从实现原理、匹配准则和搜索算法等几方面对进行了分析。
Motion object detection and tracking in serial images is the main research field in Video Surveillance, which has been widely applied in military, medicine and scientific research etc. The accuracy and stability of tracking effect depend on the design of algorithms to a great extent. The key technologies is studied in this paper about how to detect and track moving object in Video Surveillance.
On the research of the motion detection, firstly three main algorithms of motion detection and its analyzed advantages and drawbacks are researched. Secondly, background estimation based on block is proposed to solve the problem of Obtaining and estimating and updating background model which can enhance the robust of moving object detection. A background modeling and subtraction method by codebook construction is proposed. The CB algorithm adopts a quantization and clustering technique to construct a background model and subtract the current image from the background model to detect moving objects by color distortion and bright distortion. Evaluation shows that this algorithm is effective.
On the research of object tracking, firstly some common used algorithms of are researched. Then a new method of moving object tracking based on the result of image subtraction and object features matching is presented. The obtained object is framed and its center of gravity is calculated. In this together with its shape feather, its moving path can be obtained. Evaluation shows that this algorithm has better adoption in case of a little shape change and is effective in simple cases. At last the Mean Shift Algorithm is analyzed from the aspects of realization principle and match criterion and search algorithm.
引文
[1]代科学,李国辉,涂丹等.监控视频运动目标检测减背景技术的研究现状和展望[J].中国图象图形学报,2006,07:919-927.
[2]方帅,迟健男,徐心和.视频监控中运动目标跟踪算法[J].控制与决策.2005,12.
[3]艾金慰,刘克.数字视频监控系统中实时运动检测系统的研究、计算机应用研究.2005.9.
[4]R Collins,A Lipton et al.A system for video surveillance and monitoring:VSAM final report.CMU-RI-TR-00-I2,Robotic Institute Carnegie Mellon University,2000.
[5]R.Collins,A.Lipton,and T.Kaasade.Introduction to the special section on video surveillance IEEE Trans.Patern Analysis and Machine Intel Iigence.22(8):745-746,August 2000.
[6]Remagnino P,Tan T,Baker K.Muf(?)i-agent visual surveillance of dynamic scenes.Image and Vision Computing,1998,16(8):529-532.
[7]郑江滨.视频监控方法研究[D].西北工.此大学博士学位论文.
[8]龚声蓉,刘纯平,王强.数字图像处理与分析[M].北京:清华大学出版社,2006.
[9]Mallat S G.A Wavelet Tour of Signal Processing San Diego:Academic Press,1998:chapter 6.
[10]贾云得.机器视觉[M].北京:科学出版社,2000.
[11]崔宇巍.运动目标检测与跟踪中有关问题的研究[D].硕士论文,西北大学电路与系统专业,2005,5.
[12]王琳.视频运动目标跟踪中有关问题的研究[D].硕士论文,西北大学电路与系统专业,2006,5
[13]孙圣和,陆哲明.矢量量化技术及应用[M].北京:科学出版社,2002.
[14]赵燕伟,王万良.基于聚类分析的色彩量化新算法及其应用[J].计算机辅助设计与图形学学报.Vol.12,No.5,340-343
[15]王妹,卿粼波,腾奇志,何小海.矢量量化编码的码书设计研究[J].成都信息工程学院学报.2006,12:803-805
[16]K.Kim,T.H.Chalidabhongse,D.Harwood and L.Davis.Background Modeling and Subtraction by Codebook Construction[J].IEEE International Conference on Image Processing(ICIP)2004.
[17]K.Kim,T.H.Chalidabhongse,D.Harwood and L.Davis.Real-time Foreground-Background Segmentation using Codebook Model[J].Real-time Imaging,Volume 11,Issue 3,Pages 167-256,June 2005.
[18]Kyungnam Kim,D.Harwood and Larry S.Davis.Background Updating for Visual Surveillance[J].International Symposium on Visual Computing(ISVC),LNCS 3804,2005.
[19]禹晶,段娟,苏开娜.基于色度偏差的运动目标检测[J].计算机工程,2006,06:218-220.
[20]李庆忠,刘怀强,侯永海等.视频序列中运动目标自动提取的研究[J].微计算机信息,2006,10:243-244,221.
[21]李忠武,高广珠,余理富等.图像序列目标检测中阴影的消除[J].计算机应用研究,2004,05:205-206.
[22]林洪文,涂丹,李国辉.基于统计背景模型的运动目标检测方法[J].计算机工程,2003,16:97-99,101.
[23]林开颜,吴军辉,徐立鸿.彩色图像分割方法综述[J].中国图象图形学报,2005,01:1-10.
[24]Greiffenhagen M,Ramesh V,Comaniciu D,Niemann H.Statistical modeling and performance characterization of a realtime dual camera surveillance system.[J]Proceedings of International Conference on Computer Vision and Pattern Recognition 2000,2:335 42.
[25]万缨,韩毅,卢汉清.运动目标检测算法的探讨[J].计算机仿真,2006,10:221-226.
[26]赵艳启.运动目标识别与跟踪算法的研究[D].硕士论文,河海大学机械电子工程专业,2005,3.
[27]Byemer D,Mc Lauchlan P,Coifman Malikj.A real time computer vision system for measuring traffic parameters[J].IEEE Conf.Computer Vision and PatternRecogntion.1997,6:459-501.
[28]Liu 丫 Huang T S.Determining Straight Line Correspondences Intensity Image[J].Pattern Recognition.1991,24.
[29]陈康,成新,周朝辉.跟踪空中机动目标的有向图切换算法[J].火控雷达技术,2003,3.
[30]周志宇,汪亚明,黄文清基于动态图像序列的运动目标跟踪[J].浙江工程学院学报,2002,9.
[31]徐瑞鑫,刘伟宁.基于切分模板的实时跟踪算法[J].吉林工程学院学报,2002,9.
[32]张根耀,李竹林,赵宗涛.遮挡情况下运动目标的跟踪[J].安徽大学学报,2003,9.
[33]徐成华,王蕴红,谭铁牛.三维人脸建模与应用[J].中国图像图形学报,2004,9(8):893-903.
[34]刘红毅,王蕴红,谭铁牛.基于改进ENN算法的多生物特征融合的身份验证[J].自动化学报,2004,30(1),78-85.
[35]王亮,胡卫明,谭铁牛.基于步态的身份识别[J],计算机学报,2003,26(3):353-360.
[36]王亮,胡卫明,谭铁牛.人运动的视觉分析综述[J],计算机学报,2002,25(3):225-237.
[37]田园,谭铁牛,孙洪赞.一种具有良好鲁棒性的实时跟踪方法[J],计算机学报,2002,28(5):851-853.
[38]Huttenlocher D P Rucldidge W J Rucklidge W J.Comparing images using the Hausdorff distance[J].IEEE Transactions on Pattern Analysis and Machine Intelligence.1993,15:850-863.
[39]Elgammal A,Harwood D,Davis L.Non2parametric model for background subtraction[A].In:Proceedings of International Conference on Computer Vision[C],Kerkyra,Greece,1999:751-767.
[40]Cheng,Y."Mean shift,mode seeking,and clustering[J],IEEE Trans.Patern Analysis and Machine Intelligence,17:790-799,1995.
[2]方帅,迟健男,徐心和.视频监控中运动目标跟踪算法[J].控制与决策.2005,12.
[3]艾金慰,刘克.数字视频监控系统中实时运动检测系统的研究、计算机应用研究.2005.9.
[4]R Collins,A Lipton et al.A system for video surveillance and monitoring:VSAM final report.CMU-RI-TR-00-I2,Robotic Institute Carnegie Mellon University,2000.
[5]R.Collins,A.Lipton,and T.Kaasade.Introduction to the special section on video surveillance IEEE Trans.Patern Analysis and Machine Intel Iigence.22(8):745-746,August 2000.
[6]Remagnino P,Tan T,Baker K.Muf(?)i-agent visual surveillance of dynamic scenes.Image and Vision Computing,1998,16(8):529-532.
[7]郑江滨.视频监控方法研究[D].西北工.此大学博士学位论文.
[8]龚声蓉,刘纯平,王强.数字图像处理与分析[M].北京:清华大学出版社,2006.
[9]Mallat S G.A Wavelet Tour of Signal Processing San Diego:Academic Press,1998:chapter 6.
[10]贾云得.机器视觉[M].北京:科学出版社,2000.
[11]崔宇巍.运动目标检测与跟踪中有关问题的研究[D].硕士论文,西北大学电路与系统专业,2005,5.
[12]王琳.视频运动目标跟踪中有关问题的研究[D].硕士论文,西北大学电路与系统专业,2006,5
[13]孙圣和,陆哲明.矢量量化技术及应用[M].北京:科学出版社,2002.
[14]赵燕伟,王万良.基于聚类分析的色彩量化新算法及其应用[J].计算机辅助设计与图形学学报.Vol.12,No.5,340-343
[15]王妹,卿粼波,腾奇志,何小海.矢量量化编码的码书设计研究[J].成都信息工程学院学报.2006,12:803-805
[16]K.Kim,T.H.Chalidabhongse,D.Harwood and L.Davis.Background Modeling and Subtraction by Codebook Construction[J].IEEE International Conference on Image Processing(ICIP)2004.
[17]K.Kim,T.H.Chalidabhongse,D.Harwood and L.Davis.Real-time Foreground-Background Segmentation using Codebook Model[J].Real-time Imaging,Volume 11,Issue 3,Pages 167-256,June 2005.
[18]Kyungnam Kim,D.Harwood and Larry S.Davis.Background Updating for Visual Surveillance[J].International Symposium on Visual Computing(ISVC),LNCS 3804,2005.
[19]禹晶,段娟,苏开娜.基于色度偏差的运动目标检测[J].计算机工程,2006,06:218-220.
[20]李庆忠,刘怀强,侯永海等.视频序列中运动目标自动提取的研究[J].微计算机信息,2006,10:243-244,221.
[21]李忠武,高广珠,余理富等.图像序列目标检测中阴影的消除[J].计算机应用研究,2004,05:205-206.
[22]林洪文,涂丹,李国辉.基于统计背景模型的运动目标检测方法[J].计算机工程,2003,16:97-99,101.
[23]林开颜,吴军辉,徐立鸿.彩色图像分割方法综述[J].中国图象图形学报,2005,01:1-10.
[24]Greiffenhagen M,Ramesh V,Comaniciu D,Niemann H.Statistical modeling and performance characterization of a realtime dual camera surveillance system.[J]Proceedings of International Conference on Computer Vision and Pattern Recognition 2000,2:335 42.
[25]万缨,韩毅,卢汉清.运动目标检测算法的探讨[J].计算机仿真,2006,10:221-226.
[26]赵艳启.运动目标识别与跟踪算法的研究[D].硕士论文,河海大学机械电子工程专业,2005,3.
[27]Byemer D,Mc Lauchlan P,Coifman Malikj.A real time computer vision system for measuring traffic parameters[J].IEEE Conf.Computer Vision and PatternRecogntion.1997,6:459-501.
[28]Liu 丫 Huang T S.Determining Straight Line Correspondences Intensity Image[J].Pattern Recognition.1991,24.
[29]陈康,成新,周朝辉.跟踪空中机动目标的有向图切换算法[J].火控雷达技术,2003,3.
[30]周志宇,汪亚明,黄文清基于动态图像序列的运动目标跟踪[J].浙江工程学院学报,2002,9.
[31]徐瑞鑫,刘伟宁.基于切分模板的实时跟踪算法[J].吉林工程学院学报,2002,9.
[32]张根耀,李竹林,赵宗涛.遮挡情况下运动目标的跟踪[J].安徽大学学报,2003,9.
[33]徐成华,王蕴红,谭铁牛.三维人脸建模与应用[J].中国图像图形学报,2004,9(8):893-903.
[34]刘红毅,王蕴红,谭铁牛.基于改进ENN算法的多生物特征融合的身份验证[J].自动化学报,2004,30(1),78-85.
[35]王亮,胡卫明,谭铁牛.基于步态的身份识别[J],计算机学报,2003,26(3):353-360.
[36]王亮,胡卫明,谭铁牛.人运动的视觉分析综述[J],计算机学报,2002,25(3):225-237.
[37]田园,谭铁牛,孙洪赞.一种具有良好鲁棒性的实时跟踪方法[J],计算机学报,2002,28(5):851-853.
[38]Huttenlocher D P Rucldidge W J Rucklidge W J.Comparing images using the Hausdorff distance[J].IEEE Transactions on Pattern Analysis and Machine Intelligence.1993,15:850-863.
[39]Elgammal A,Harwood D,Davis L.Non2parametric model for background subtraction[A].In:Proceedings of International Conference on Computer Vision[C],Kerkyra,Greece,1999:751-767.
[40]Cheng,Y."Mean shift,mode seeking,and clustering[J],IEEE Trans.Patern Analysis and Machine Intelligence,17:790-799,1995.