智能视频监控中的目标检测与跟踪技术
|
摘要
智能化视频监控系统是以数字化、网络化视频监控为基础,它利用图像处理、模式识别等技术,抽取并分析视频源中的关键信息,及时发现并处理监控场景下的异常情况,从而更加有效的协助安全人员处理危机,并最大限度的降低误报和漏报现象。在我国构建社会主义和谐社会不断推进,“建设平安城市,促进社会和谐”已经成为全社会共同关注的目标的大环境下,智能视频监控技术在科学研究和工程应用上将会有着非常广阔的前景和巨大的潜在经济价值。
运动目标检测与跟踪技术作为智能视频监控的一个重要组成部分近年来取得了长足的发展。但在实现可靠的实际应用前,还需要解决许多相关难题,如复杂环境下的目标检测,遮挡情况下的目标跟踪,以及目标跟踪的实时性等。本文在对目标检测和跟踪领域的主流算法分析研究的基础上,针对前两个问题提出了一些行之有效的解决方法,主要研究成果如下:
运动目标检测方面:编程实现了主流的运动目标检测算法,包括帧差法、背景差法,光流法,分析了它们各自的优势和不足;在对现有的高斯背景建模方法进行深入分析的基础上,提出了一种改进的高斯背景建模方法。通过大量的实验验证表明,利用本文提出的改进算法进行目标检测具有较高的准确率,能够有效地检测出背景中运动目标的完整区域,能够满足实时性检测的需要。
运动目标跟踪方面:主要研究了Mean Shift算法和Kalman滤波在目标跟踪领域中的应用,鉴于二者各有优缺点且可优势互补,本文提出了将Kalman预测和Mean shift搜索相结合的运动目标跟踪新方法。利用Kalman滤波估计出运动目标在下一帧中最可能出现的位置,再用Mean shift法据此进行较小范围的搜索和目标匹配,减少颜色分布对目标跟踪的影响,用较小的运算量获得较为可靠的跟踪效果。对于遮挡情况,利用Mean Shift中的相似度函数来判断,若发生遮挡,不再用Mean Shift收敛点信息去更新Kalman滤波器,直接使用预测点信息去更新,有效解决了目标遮挡问题。
Intelligent video surveillance system is based on digital, networked surveillance, which can extract and analyze the key information from video source, timely detect and deal with the unusual cases on the monitored scene, thus to be able to effectively assist the security personnel dealing with the crisis and minimizing false negative and false positive by using image processing and pattern recognition techniques. When the building of a HharmoniousH HsocietyH is emphasized in our country and“building safe city, and promoting social harmony”is becoming a society-wide goal, the intelligent video surveillance technology has broad application prospect and great potential economic value both in scientific research and engineering applications.
As a critical part of intelligent video surveillance system, detection and tracking of moving targets is becoming a hotspot in computer vision community. However, before it get an extensive and reliable applications, several problems have to be resolved such as object detection in complex scenario, tracking in occlusion condition and real time tracking. On the basis of the analysis of dominant algorithms for target detection and tracking, we propose several techniques for solving the first two problems. The main contributions of this thesis are as follows:
Algorithms for object detection are studied. Some classical algorithms for moving target detection, including HframeH differential method, background subtraction method, optical flow method are implemented, their strengths and weaknesses are discussed; After currently widely used Gaussian background modeling is discussed in detail, an improved method for Gaussian background modeling was proposed. The experiment results show that our improved method can extract effectively the moving objects and fit for real time detection.
Algorithms for object tracking are studied. Mean shift algorithm and Kalman filter is applied in object tracking. Considering each of them has its own disadvantages and they have complementary advantages, we present a new algorithm combining the Kalman prediction with mean shift. By using Kalman filter to predict locations where moving objects most probably appear in the next frame and mean shift to search in the corresponding areas and match the moving objects, the approach promises to obtain more reliable tracking effect with much less computation cost. We determine whether the tracking target is occluded by similarity function in Mean Shift. When the occlusion happens, we update filter by using directly predicting points rather than the convergence point information of mean shift. The experiment results show that our algorithm can track the moving target well and also has better robustness under occlusion.
运动目标检测与跟踪技术作为智能视频监控的一个重要组成部分近年来取得了长足的发展。但在实现可靠的实际应用前,还需要解决许多相关难题,如复杂环境下的目标检测,遮挡情况下的目标跟踪,以及目标跟踪的实时性等。本文在对目标检测和跟踪领域的主流算法分析研究的基础上,针对前两个问题提出了一些行之有效的解决方法,主要研究成果如下:
运动目标检测方面:编程实现了主流的运动目标检测算法,包括帧差法、背景差法,光流法,分析了它们各自的优势和不足;在对现有的高斯背景建模方法进行深入分析的基础上,提出了一种改进的高斯背景建模方法。通过大量的实验验证表明,利用本文提出的改进算法进行目标检测具有较高的准确率,能够有效地检测出背景中运动目标的完整区域,能够满足实时性检测的需要。
运动目标跟踪方面:主要研究了Mean Shift算法和Kalman滤波在目标跟踪领域中的应用,鉴于二者各有优缺点且可优势互补,本文提出了将Kalman预测和Mean shift搜索相结合的运动目标跟踪新方法。利用Kalman滤波估计出运动目标在下一帧中最可能出现的位置,再用Mean shift法据此进行较小范围的搜索和目标匹配,减少颜色分布对目标跟踪的影响,用较小的运算量获得较为可靠的跟踪效果。对于遮挡情况,利用Mean Shift中的相似度函数来判断,若发生遮挡,不再用Mean Shift收敛点信息去更新Kalman滤波器,直接使用预测点信息去更新,有效解决了目标遮挡问题。
Intelligent video surveillance system is based on digital, networked surveillance, which can extract and analyze the key information from video source, timely detect and deal with the unusual cases on the monitored scene, thus to be able to effectively assist the security personnel dealing with the crisis and minimizing false negative and false positive by using image processing and pattern recognition techniques. When the building of a HharmoniousH HsocietyH is emphasized in our country and“building safe city, and promoting social harmony”is becoming a society-wide goal, the intelligent video surveillance technology has broad application prospect and great potential economic value both in scientific research and engineering applications.
As a critical part of intelligent video surveillance system, detection and tracking of moving targets is becoming a hotspot in computer vision community. However, before it get an extensive and reliable applications, several problems have to be resolved such as object detection in complex scenario, tracking in occlusion condition and real time tracking. On the basis of the analysis of dominant algorithms for target detection and tracking, we propose several techniques for solving the first two problems. The main contributions of this thesis are as follows:
Algorithms for object detection are studied. Some classical algorithms for moving target detection, including HframeH differential method, background subtraction method, optical flow method are implemented, their strengths and weaknesses are discussed; After currently widely used Gaussian background modeling is discussed in detail, an improved method for Gaussian background modeling was proposed. The experiment results show that our improved method can extract effectively the moving objects and fit for real time detection.
Algorithms for object tracking are studied. Mean shift algorithm and Kalman filter is applied in object tracking. Considering each of them has its own disadvantages and they have complementary advantages, we present a new algorithm combining the Kalman prediction with mean shift. By using Kalman filter to predict locations where moving objects most probably appear in the next frame and mean shift to search in the corresponding areas and match the moving objects, the approach promises to obtain more reliable tracking effect with much less computation cost. We determine whether the tracking target is occluded by similarity function in Mean Shift. When the occlusion happens, we update filter by using directly predicting points rather than the convergence point information of mean shift. The experiment results show that our algorithm can track the moving target well and also has better robustness under occlusion.
引文
[1] Collins R.Asystem for video surveillance and monitoring.CarNegie Mellon University Technical Report,2000.
[2]王晨.智能视频监控设计:[硕士学位论文].南京:南京理工大学,2007.
[3]陈轶博.智能视频监控系统的设计与实现:[硕士学位论文].大连:大连海事大学,2008.
[4]陈功.鲁棒的智能视频监控方法研究:[博士学位论文].合肥:中国科学技术大学,2008.
[5]谭铁牛.智能视觉监控技术概述.第一届全国智能视觉监控学术会议,2002.
[6] UHhttp://www.cbsr.ia.ac.cn/china/Surveillance%20CH.aspU
[7]万缨,韩毅,卢汉清.运动目标检测算法的探讨.计算机仿真,2006,10(10):56-62.
[8] Hu Weiming,Tan Tieniu,Wang Liang, et al.A Survey on Visual Surveillance of Object Motion and Behaviors.IEEE Transactions on System,Man,and Cybernetics,2004, 34(3):334-352.
[9]薛晨,朱明,刘春香.遮挡情况下目标跟踪算法综述.中国光学与应用光学,2009,2(5):388-394.
[10]刘泉志.基于视频的运动目标检测与跟踪算法研究[硕士学位论文].上海:上海交通大学,2009.
[11]刘国宏,郭文明.改进的中值滤波去噪算法应用分析,计算机工程与应用,2010,46(10):187-189.
[12]彭波,崔永普,吕小晴.用于数字监控系统的图像去噪算法的研究与实现.中国农业大学学报,2004, 9(5):62-66.
[13]史忠科,曹力.交通图像检测与分析.北京:科技出版社,2007.22.
[14] R.C.Gonzalez,R.E.woods.数字图像处理(第2版).北京:电子工业出版社.2003.
[15] A Neri,S Clonnese,G Russo etal.Automatic moving object and background separation. Signal Processing, 1998, 66:219-232.
[16] Haritaoglu, Land Harwood,D,Davis,etc.A real time system for detecting and tracking people. Proceedings of Third IEEE International Conference on Automatic Face and Gesture Recognition.l998:222-227.
[17] Stringa,E.and Regazzoni,C.S.Real time video shot detection for scene surveillance applications.IEEE TransactionsonImage rocessing.2000,1(1):69-79.
[18]吴晓阳.OpenCV的运动目标检测与跟踪[硕士学位论文].杭州:浙江大学,2008.
[19] Barron J,Fleet D,Beauchemin S.Performance of optical flow techniques. International journal of computer vision, 1994, 12(1):42-77.
[20]贾海涛.运动目标检测与识别算法的研究[硕士学位论文].成都:电子科技大学,2006.
[21] J.Barron, D.Fleet,and S.Beauchemin,“Performance of Optical Flow Techniques”Inter- national Journal of Computer Vision, vol. 12, no. 1, pp. 42-77,1994.
[22] C.R.Wren, A.Azarbayejani, T.Darrcll et al. Real-time tracking of the human body. IEEE Transactions on pattern analysis and machine intelligence,1997,19(7): 780-785.
[23] Friedman N,Russell S.Image segmentation in video se-quences:a probabilistic approach.In: Proceedings of Thirteenth Conference on Uncertainty in Articial Intelligence. Providence, USA.1997.
[24] Stauffer C,Grimson W.Adaptive background mixture models for realtime tracking. In: Proceedings of IEEE International Conference on Computer Vision and Pattern Recognition, Fort Collies,Colorado,USA,1999,2:246-2.
[25] KaewTraKulPong P, Bowden R. An improved adaptive background mixture model for realtime tracking with shadow detection. In: Proceedings of 2nd European Workshop Advanced VideoBased Surveillance Systems. Providence, Kluwer Academic Publishers. 2001.
[26] Zivkovic Z,van der Heijden F. Efficient adaptive density estimation per image pixel for the task of background subtraction. Pattern Recognition Letters, 2006, 27: 773-780
[27] Lee D S.Effective Gaussian mixture learning for video background subtraction. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2005, 27(5): 827-832.
[28]陈璇,吴清江.基于色度坐标高斯混合模型的步态检测.计算机工程,2009,35(17):198-200.
[29]吕国亮,赵曙光,赵俊.基于三帧差分和连通性检验的图像运动目标检测新方法.液晶与显示,2007(1):87-92.
[30]左军毅,梁彦,赵春晖,等.Mean-Shift跟踪算法中尺度自适应策略的研究.中国图象图形学报,2008, 13(9):1750-1757.
[31] LIUPR,MENGM Q H,LIUPX,et al.Optical flow and active contour for moving object segmentation and detection in monocular robot.Proceedings 2006 IEEE International Conference on Robotics and Automation. Washington, DC:IEEE,2006: 4075-4080.
[32]曾伟,朱桂斌,李瑶.基于Kalman点匹配估计的运动目标跟踪.计算机应用,2009,29(6):1677-1682.
[33]罗嘉,韦志辉.基于几何活动轮廓模型的目标跟踪与快速运动估计.中国图象图形学报,2009,14 (7):1361-1368.
[34] COMANICIU D,RAMESH V,MEER P.Real-time tracking of nonrigid objects using Mean-Shift.IEEE Computer Vision and Pattern Recognition.Washington,DC: IEEE,2000:142-149.
[35] K.FUKUNAGE,L.D.HOSTETLER. The estimation of the gradient of a density function with application in recognition.IEEE Trans.on Information Theory,1975,21(1):32-40.
[36] Y.CHENG.Mean shift,mode seeking and clustering.IEEE Trans.on Pattern Analysis and Machine Intelligence,1995,17(8):790-799.
[37] COMANICIU D,RAMESH V,MEER P.Real time tracking of nonrigid objects using Mean-Shift.IEEE Computer Vision and Pattern Recognition.Washington,DC: IEEE,2000:142 - 149.
[38]朱胜利,朱善安,李旭超.快速运动目标的Mean shift跟踪算法.光电工程,2006,33(5):66-70.
[39] Simon Haykin.自适应滤波器原理.郑宝玉译.电子工业出版社.2003.7
[40]董士崇,王天珍,许刚.视频图像中的运动检测.武汉理工大学学报.信息与管理工程版2004Aug. Vo1.26.No.4.
[41]杨戈,刘宏.视觉跟踪算法综述.智能系统学报.2010,5(2):95-103.
[42] Uhttp://www.research.ibm.com/peoplevision/performanceevaluation.html
[43]刘瑞祯,于仕琪.OpenCV教程一基础篇.北京航空航天大学出版社,2007.
[2]王晨.智能视频监控设计:[硕士学位论文].南京:南京理工大学,2007.
[3]陈轶博.智能视频监控系统的设计与实现:[硕士学位论文].大连:大连海事大学,2008.
[4]陈功.鲁棒的智能视频监控方法研究:[博士学位论文].合肥:中国科学技术大学,2008.
[5]谭铁牛.智能视觉监控技术概述.第一届全国智能视觉监控学术会议,2002.
[6] UHhttp://www.cbsr.ia.ac.cn/china/Surveillance%20CH.aspU
[7]万缨,韩毅,卢汉清.运动目标检测算法的探讨.计算机仿真,2006,10(10):56-62.
[8] Hu Weiming,Tan Tieniu,Wang Liang, et al.A Survey on Visual Surveillance of Object Motion and Behaviors.IEEE Transactions on System,Man,and Cybernetics,2004, 34(3):334-352.
[9]薛晨,朱明,刘春香.遮挡情况下目标跟踪算法综述.中国光学与应用光学,2009,2(5):388-394.
[10]刘泉志.基于视频的运动目标检测与跟踪算法研究[硕士学位论文].上海:上海交通大学,2009.
[11]刘国宏,郭文明.改进的中值滤波去噪算法应用分析,计算机工程与应用,2010,46(10):187-189.
[12]彭波,崔永普,吕小晴.用于数字监控系统的图像去噪算法的研究与实现.中国农业大学学报,2004, 9(5):62-66.
[13]史忠科,曹力.交通图像检测与分析.北京:科技出版社,2007.22.
[14] R.C.Gonzalez,R.E.woods.数字图像处理(第2版).北京:电子工业出版社.2003.
[15] A Neri,S Clonnese,G Russo etal.Automatic moving object and background separation. Signal Processing, 1998, 66:219-232.
[16] Haritaoglu, Land Harwood,D,Davis,etc.A real time system for detecting and tracking people. Proceedings of Third IEEE International Conference on Automatic Face and Gesture Recognition.l998:222-227.
[17] Stringa,E.and Regazzoni,C.S.Real time video shot detection for scene surveillance applications.IEEE TransactionsonImage rocessing.2000,1(1):69-79.
[18]吴晓阳.OpenCV的运动目标检测与跟踪[硕士学位论文].杭州:浙江大学,2008.
[19] Barron J,Fleet D,Beauchemin S.Performance of optical flow techniques. International journal of computer vision, 1994, 12(1):42-77.
[20]贾海涛.运动目标检测与识别算法的研究[硕士学位论文].成都:电子科技大学,2006.
[21] J.Barron, D.Fleet,and S.Beauchemin,“Performance of Optical Flow Techniques”Inter- national Journal of Computer Vision, vol. 12, no. 1, pp. 42-77,1994.
[22] C.R.Wren, A.Azarbayejani, T.Darrcll et al. Real-time tracking of the human body. IEEE Transactions on pattern analysis and machine intelligence,1997,19(7): 780-785.
[23] Friedman N,Russell S.Image segmentation in video se-quences:a probabilistic approach.In: Proceedings of Thirteenth Conference on Uncertainty in Articial Intelligence. Providence, USA.1997.
[24] Stauffer C,Grimson W.Adaptive background mixture models for realtime tracking. In: Proceedings of IEEE International Conference on Computer Vision and Pattern Recognition, Fort Collies,Colorado,USA,1999,2:246-2.
[25] KaewTraKulPong P, Bowden R. An improved adaptive background mixture model for realtime tracking with shadow detection. In: Proceedings of 2nd European Workshop Advanced VideoBased Surveillance Systems. Providence, Kluwer Academic Publishers. 2001.
[26] Zivkovic Z,van der Heijden F. Efficient adaptive density estimation per image pixel for the task of background subtraction. Pattern Recognition Letters, 2006, 27: 773-780
[27] Lee D S.Effective Gaussian mixture learning for video background subtraction. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2005, 27(5): 827-832.
[28]陈璇,吴清江.基于色度坐标高斯混合模型的步态检测.计算机工程,2009,35(17):198-200.
[29]吕国亮,赵曙光,赵俊.基于三帧差分和连通性检验的图像运动目标检测新方法.液晶与显示,2007(1):87-92.
[30]左军毅,梁彦,赵春晖,等.Mean-Shift跟踪算法中尺度自适应策略的研究.中国图象图形学报,2008, 13(9):1750-1757.
[31] LIUPR,MENGM Q H,LIUPX,et al.Optical flow and active contour for moving object segmentation and detection in monocular robot.Proceedings 2006 IEEE International Conference on Robotics and Automation. Washington, DC:IEEE,2006: 4075-4080.
[32]曾伟,朱桂斌,李瑶.基于Kalman点匹配估计的运动目标跟踪.计算机应用,2009,29(6):1677-1682.
[33]罗嘉,韦志辉.基于几何活动轮廓模型的目标跟踪与快速运动估计.中国图象图形学报,2009,14 (7):1361-1368.
[34] COMANICIU D,RAMESH V,MEER P.Real-time tracking of nonrigid objects using Mean-Shift.IEEE Computer Vision and Pattern Recognition.Washington,DC: IEEE,2000:142-149.
[35] K.FUKUNAGE,L.D.HOSTETLER. The estimation of the gradient of a density function with application in recognition.IEEE Trans.on Information Theory,1975,21(1):32-40.
[36] Y.CHENG.Mean shift,mode seeking and clustering.IEEE Trans.on Pattern Analysis and Machine Intelligence,1995,17(8):790-799.
[37] COMANICIU D,RAMESH V,MEER P.Real time tracking of nonrigid objects using Mean-Shift.IEEE Computer Vision and Pattern Recognition.Washington,DC: IEEE,2000:142 - 149.
[38]朱胜利,朱善安,李旭超.快速运动目标的Mean shift跟踪算法.光电工程,2006,33(5):66-70.
[39] Simon Haykin.自适应滤波器原理.郑宝玉译.电子工业出版社.2003.7
[40]董士崇,王天珍,许刚.视频图像中的运动检测.武汉理工大学学报.信息与管理工程版2004Aug. Vo1.26.No.4.
[41]杨戈,刘宏.视觉跟踪算法综述.智能系统学报.2010,5(2):95-103.
[42] Uhttp://www.research.ibm.com/peoplevision/performanceevaluation.html
[43]刘瑞祯,于仕琪.OpenCV教程一基础篇.北京航空航天大学出版社,2007.