摘要
视频监控系统的智能化是计算机视觉领域新兴的一个研究方向。它的主要目标是通过计算机视觉技术对监控视频序列的内容进行自动分析和判断,对监控过程中出现的异常行为及时做出反应。视频运动目标的检测与跟踪技术是智能视频监控系统中的关键底层技术,它涉及数字图像处理、机器学习、人工智能等众多技术领域。目标检测与跟踪技术的研究对安防监控、交通监控等应用领域有重要的现实意义。
本文重点研究智能视频监控系统中的运动目标检测和目标跟踪两项关键技术。在分析对比常用的运动目标检测与跟踪方法的基础上,结合实际应用对算法进行了一定的研究和改进。具体研究工作如下：
(1)运动目标检测方面,首先概要介绍了背景减除法、帧间差分法和光流法,并分析了它们各自的优缺点；在此基础上,着重研究了基于混合高斯模型的背景减除法；通过对经典混合高斯模型的建立和参数更新进行数学描述,指出经典混合高斯模型存在的计算量大、实时性差的缺点；考虑到实际应用场景中常出现的阴影干扰问题,本文提出了一种YUV(亮度,色度)彩色模型下的基于混合高斯阴影模型的阴影抑制算法,通过实验验证,该阴影抑制算法一定程度上消除了阴影对运动目标检测准确性的影响；并针对混合高斯模型计算量大的问题提出了一种改进的混合高斯模型快速算法,实验表明,该算法有效提升了运动目标检测的效率。
(2)目标跟踪方面,首先分析对比了基于滤波理论、基于偏微分方程和基于均值平移的目标跟踪方法的优缺点及适用条件；重点研究了基于均值平移算法的目标跟踪方法,分析了该方法易于实现、算法效率高等优点以及由于特征不足而造成跟踪鲁棒性差的缺点；针对经典均值平移算法的缺点,在详细描述了尺度不变特征转换算法的原理及优势的基础上,提出了一种结合尺度不变特征转换算法与均值平移跟踪算法的融合算法,并通过实验验证了该融合算法相较于经典均值平移算法在目标跟踪鲁棒性方面的提升。
(3)介绍了智能视频监控系统的应用及设计架构,设计并实现了该系统中行人移动侦测与跟踪子系统、绊线检测与周界防范子系统和盗移检测子系统；对各子系统的运行效果进行了实验和分析,验证了本文所研究的运动目标检测与目标跟踪算法在实际应用中是适用的。
Intelligentization of video surveillance system is an emerging research orientation in the field of computer vision. Its main goal is to realize the automatic analysis and judgment of the surveillance video sequences by computer vision technologies, to respond to the occurrence of the abnormal behavior under surveillance without delay. The object detection and tracking are the key technologies for intelligent video surveillance system, and they concern many fields such as digital image processing, machine learning, and artificial intelligence, etc. The research of object detection and tracking technologies has important practical significance in the field of security monitoring, traffic monitoring, and so on.
This paper is committed to the key issues of moving object detection and tracking in intelligent video surveillance systems. On the basis of analysis and comparison of the object detection and tracking methods, and combined with the practical application, certain research and improvement of the methods are taken. The main research content is as follows:
(1) In terms of moving objects detection, background subtraction, temporal differencing and optical flow are introduced briefly, and the advantages and disadvantages of the three methods are analyzed. On this basis, the background subtraction method based on Gaussian Mixture Model (GMM) is focused on. By describing the classic Gaussian Mixture Model and its parameter updating mathematically, point out that the classic Gaussian Mixture Model has the defects of high computational complexity and poor real-time performance. Taking into account that the shadow interference often occurs in practical application scenarios, a shadow suppression algorithm based on Gaussian Mixture Shadow Model (GMSM) in YUV (Luminance, Chrominance) color space is proposed. Proved by experiments, to some degree, this shadow suppression algorithm could eliminate some influence to the object detection accuracy caused by shadow.To solve the problems of high computational complexity, an improved fast algorithm for Gaussian Mixture Model is proposed, and experiments show that this improved algorithm is effective to enhance the efficiency of moving objects detection.
(2) In terms of object tracking, the advantages and disadvantages, and applicable conditions of the tracking methods based on Filtering Theory, Partial Differential Equations and Mean Shift respectively, are analyzed. The tracking method based on Mean Shift is focused on, and the advantages such as being easy to implement and high efficiency are analyzed, as well as the disadvantages such as bad robustness caused by lack of characteristics. Against the shortcoming of the classic Mean Shift algorithm, after detailed analysis of the principle and advantages of Scale Invariant Feature Transform (SIFT) algorithm is made, a fusion algorithm combining SIFT and classic Mean Shift is proposes, followed by some experiments which prove that the fusion algorithm could effectively improve the robustness in object tracking, compared to the classic Mean Shift algorithm.
(3) The application and design architecture of intelligent video surveillance systems are introduced in detail, and three subsystems, namely Pedestrian Detection and Tacking, Tripwire Detection&Regional Preparedness, and Theft Detection, are designed and implemented. Experiments and analysis to the subsystems are done, verifying that the moving object detection and tracking algorithms researched in this paper are applicable in practical applications.
引文
[1]GB50348-2004,安全防范工程技术规范[S].
[2]STOR-AGE.IDC分析师解析中国视频监控市场现状与发展趋势[EB/OL]. http://www.cnetnews.com.cn/2008/0807/1045955.shtml,2008-08-07.
[3]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,2012：1-2.
[4]程丽达.电视监控系统综述[J].有线电视技术,2005,15：5-8.
[5]乔彩凤,宋世军,何忠.数字视频监控系统的智能化实现[J].计算机与现代化,2007,12：45-48,51.
[6]高俊祥.智能视频监控中目标的检测与跟踪[学位论文].北京邮电大学,2010.
[7]Carnegie Mellon University. VSAM DEMO'99[EB/OL]. http://www.cs.cmu.edu/-vsam/OldVsamWeb/vsam99dem o.html,1999-10-19.
[8]赵春晖,潘泉,梁彦,等.视频图像运动目标分析[M].北京：国防工业出版社,2011.
[9]Wren C, Azarbayejani A, Darrell T, et al. Pfinder:Real-time tracking of the human body[J]. IEEE Trans. on PAMI,1997,19(7):780-785.
[10]Stauffer C, Grimson W. Adaptive background mixture models for real-time tracking[C]. CVPR,1999,2:246-252.
[11]Kaew TraKulPong P, Bowden R. An Improved Adaptive Background Mixture Model for Realtime Tracking with Shadow Detection[C].2nd European Workshop on Advanced Video Based Surveillance Systems,2001,5308:1-5.
[12]Elgammal A, Duraiswam R, Davis L S. Efficient Kernel Density Estimation Using the Fast Gauss Transform with Applications to Segmentation and Tracking[C]. ICCV,2001,25(11):1449-1504.
[13]Gordon N J, Salmond D J, Smith A F M. Novel approach to nonlinear/ non-Gaussian Bayesian state estimation[J]. IEE Proceedings F,1993,140(2): 107-113.
[14]Isard M, Blake A. Contour tracking by stochastic propagation of conditional density[C]. Proc. European Conf. Computer Vision, Berlin:Springer-Verlag, 1996:343-356.
[15]Isard M, Blake A. Icondensation:Unifying low-level and high-level tracking in a stochastic framework[C]. Proc. European Conf. Computer Vision, Berlin: Springer-Verlag,1998:893-908.
[16]Spengler M, Schiele B. Towards robust multi-cue integration for visual tracking[J]. Machine Vision and Applications,2003,14(1):50-58.
[17]Okuma K, Taleghani A, Nando de Freitas, et al. A boosted particle filter: Multitarget detection and tracking[C]. Proc. European Conf. Computer Vision, Berlin:Springer-Verlag,2004:28-39.
[18]Fukunaga K, Hostetler L D. The estimation of the gradient of a density function, with applications in pattern recognition[J]. IEEE Transactions on Information Theory,1975,21(1):32-87.
[19]Cheng Y. Mean Shift, mode seeking, and clustering[J]. IEEE Transactions Pattern Analysis and Machine Intelligence,1995,17(8):790-799.
[20]Comaniciu D, Ramesh V, Meer P. Kernel-Based Object Tracking[J]. IEEE Transactions on PAMI,2003,25(5):564-577.
[21]Yang C J, Duraiswami R, de Menthon D, et al. Mean-shift analysis using quasi-newton methods[C]. Proceedings of IEEE Conference on Image Processing, New York:IEEE Press,2003,:447-450.
[22]David G. Lowe. Distinctive Image Features from Scale-Invariant Keypoints[J]. International Journal of Computer Vision,2004,60(2):91-110.
[23]Ai-hua Chen, Ming Zhu, Yan-hua Wang, et al. Mean Shift Tracking Combining SIFT[C]. ICSP 2008:9th International Conference on Signal Processing,2008:1532-1535.
[24]PETS2006[EB/OL]. http://ftp.pets.rdg.ac.uk/PETS2006/.
[25]Rafael C. Gonzalez, Richard E. Woods. Digital Image Processing, Third Edition[M]. New Jersey:Prentice Hall,2008. [26]WIKIPEDIA. YUV[EB/OL]. http://en.wikipedia.org/wiki/YUV.
[27]何斌,马天矛,王运坚,等Visual C++数字图像处理,第二版[M].北京：人民邮电出版社,2002：4-8.
[28]邵丹,韩家伟.YUV与RGB之间的转换[J].长春大学学报,2004,14(4)：51-53.
[29]姚春莲,周兵.运动对象检测及其在视频压缩与处理中的应用[M].北京：冶金工业出版社,2010：28-31.
[30]Wu Q Z, Jeng B S. Background subtraction based on logarithmic intensities[J]. Pattern Recognition Letter,2002,23(13):1529.
[31]Kim K, Chalidabhongse T, Harwood D, et al. Real-time fore-ground-back-ground segmentation using codebook model[J]. Real Time Imaging,2005,11(3):172.
[32]Fermuller C, Shulman D, Aloimonos Y. The statistics of optical flow[J]. Computer Vision and Image Understanding,2001,82:1-32.
[33]J. J Gibson. The Perception of the Visual World[M]. Houghton Mifflin, Boston, MA,1950.
[34]Friedman N, Russell S. Image segmentation in video sequences:A probabilistic approach[C]. Proceedings Thirteenth Conference on Uncertainty in Artificial Intelligence,1997:175-181.
[35]Z Zivkovic. Improved Adaptive Gaussian Mixture Model for Background Subtraction[C]. International Conference on Pattern Recognition,2004,2:28-31.
[36]T. Bouwmans, F. El Baf, B. Vachon. Background Modeling using Mixture of Gaussians for Foreground Detection-A Survey [J]. Recent Patents on Computer Science,2008,1(3):219-237.
[37]Martel-Brisson N, Zaccarin A. Moving cast shadow detection from a Gaussian mixture shadow model [C]. IEEE Conference on Computer Vision and Pattern Recognition,2005,2:643-648.
[38]王典,程咏梅,杨涛,等.基于混合高斯模型的运动阴影抑制算法[J].计算机应用,2006,26(5)：1021-1023,1026.
[39]Chen B S, Lei Y. Indoor and outdoor people detection and shadow suppression by exploiting HSV color information[C]. International Conference on Computer and Information Technology,2004:137-142.
[40]曹玉东.图像检索中的特征表示和索引方法的研究[学位论文].北京邮电大学,2011.
[41]Alper Yilmaz. Object Tracking:A Survey[J]. ACM Computing Surveys, 2006,38(4):13-es.
[42]李培华.序列图像中运动目标跟踪方法[M].北京：科学出版社,2010：9-16.
[43]Li P H, Zhang T W. Pece A E C. Visual contour tracking based on particle filters[J]. Image and Vision Computing,2003,21(1):111-123.
[44]Kass M, Witkins A, Terzopoulos D. Snakes:Active contour models[J]. International Journal of Computer Vision,1988,1(4):321-331.
[45]Chan T F, Vese L A. Active contours without edges[J]. IEEE Transactions on Image Processing,2001,10(2):266-277.
[46]Cremers D, Rousson M, Deriche R. A review of statistical approaches to level set segmentation:integrating color, texture, motion and shape[J]. International Journal of Computer Vision,2007,72(2):195-215.
[47]Aubert G, Barlaud M, Faugeras O, et al. Image segmentation using active contours:Calculus of variations or shape gradients?[J]. SIAM Journal of Applied Mathematics,2003,63(6):2128-2154.
[48]Osher S, Sethian J A. Fronts propagating with curvature-dependent speed: Algorithms based on hamilton-jacobi formulations[J]. Journal of Computational Physics,1988,79:12-49.
[49]李培华,肖莉娟.基于Mean Shift的相似性变换和仿射变换目标跟踪算法[J].中国图象图形学报,2011,16(2)：258-266.
[50]张东亚.结合改进的SIFT特征匹配方法的运动跟踪算法[学位论文].上海交通大学,2010.
[51]Kai Du, YongfengJu, YinliJin, et al. Object tracking based on improved MeanShift and SIFT[C].2012 2nd International Conference on Consumer Electronics, Communications and Networks,2012:2716-2719.
[52]翟旭,戚玲,喻松.基于霍夫曼树SVM的监控视频清晰度评价[J].软件,2012,33(12)：100-103.
[53]刘治红,骆云志.智能视频监控技术及其在安防领域中的应用[J].兵工自动化,2009,28(4)：75-77.