无重叠视域多摄像机运动目标匹配研究
|
摘要
随着智能视频监控系统的高速发展，近年来，人们对视频监控系统的智能性要求日益提高。多摄像机网络能够监控更广阔的区域，正在逐渐被重视。但是，从经济、资源等方面的考虑，不太可能布置大量的摄像头覆盖所有监视区域。因此，无重叠视域多摄像机情况下的目标跟踪就成了广域视频监控研究的主要内容之一。无重叠视域多摄像机间的目标匹配作为跟踪的重要前提和必须步骤，也成为广域视频监控研究中的重要研究内容。
本文针对无重叠视域多摄像机监控环境中不规则扰动背景的运动目标检测问题、固定单摄像机下的目标跟踪问题和多摄像机间的目标匹配问题进行了深入研究，取得了如下成果：
1．针对混合高斯背景模型易将场景中的不规则扰动背景成分当作前景目标区域，造成误检率较高的问题，本文提出了一种融合区域运动估计的改进算法（ME-GMM）。该方法首先利用混合高斯背景模型有效应对场景中规则扰动并对前景目标区域进行预提取，然后通过区域运动估计进一步分析各区域的运动特征来区分不规则扰动背景成分和运动前景目标区域，从而完成实际监控场景的运动目标检测。在数据集I2R[58]实验结果表明，与现有算法相比，本文提出的算法能更好地检测不规则背景扰动情况下的运动目标。
2．针对固定单摄像头视域内基于单一特征的目标跟踪算法容易发生目标位置偏移、目标丢失等问题，本文提出了一种基于多特征协方差矩阵的目标跟踪算法。该算法首先利用ME-GMM运动目标检测算法提取出每一帧的运动目标区域并去除视频流中的背景成分，有效避免了跟踪过程可能发生的目标位置偏移；然后运用多特征协方差矩阵对连续帧间的目标区域进行区域匹配，并依此分析目标在连续帧间的运动状态预测目标在下一帧中可能出现的位置，有效克服了因遮挡导致的跟踪丢失问题。在数据集PETS2009[59]上的实验结果表明，与基于单一特征的粒子滤波的跟踪算法相比，本文提出的目标跟踪算法目标位置更准确，跟踪丢失率更低。
3．针对无重叠视域中目标在不同摄像间外观容易发生变化，使得目标在多摄像机间难以获得准确匹配的问题，在分析颜色不变性特征的基础上，提出了一种基于AP聚类算法的形状上下文目标表现模型。该模型首先通过AP聚类算法确定目标图像在log颜色空间的聚类中心和各像素所属的类别，然后将所有聚类中心作为参照点集合生成形状上下文描述子集合，最后通过计算无重叠视域多摄像机监控环境下获取的不同目标表现模型之间的EMD距离实现相同目标的匹配，降低了光照变化、角度变化和尺度变化所带来的匹配误差。在数据集VIPeR[60]的仿真实验证明，本文提出的目标表现模型可以有效实现无重叠视域多摄像机监控环境下的目标的匹配。
In recent years, with the rapid development of intelligent video surveillance system,as well as the urgent demand of the security situation, people’s requirements for smartintelligent video surveillance system are increasing more and more, and it is muchaccounted of day by day that multi-cameras network because of its expansive surveillanceareas. Nevertheless, on the one hand, due to the consideration of capital and resources, it isunlikely to cover all surveillance area with a great amount of monitors. Therefore, cameratracking with non-overlapping views has become a main part in the study of wide-areavideo surveillance. On the other hand, the precondition in the camera tracking is objectmatching which is the significant research project in the object detection and matching formultiple non-overlapping surveillance cameras.
This thesis contrapose the existing problem in the moving object detection ofdisturbed irregular background, object tracking with a single static camera and objectre-identification for multiple non-overlapping surveillance cameras. With the in-depthstudy, the following results can be achieved:
1. To the question the high rate of false drop of moving target detection with Gaussianmixture model in the irregular disturbed background, combined with the motion estimationmethod usually used in video coding, we proposed an effective method of real-timemoving object detection. Gaussian mixture model can effectively cope with regulardisturbance in the scene, and has an advantage of foreground area extraction, at the sametime, irregular disturbing region in the background will also be detected as movingforeground region. Therefore, with the detection results of Gaussian mixture model, wefurther utilize the regional motion estimation method to differentiate irregular disturbingregion of background and prospect moving object region. The experimental results indataset I2R[58]show that it is very effective that the method detailed in this paper that objectdetection for irregular disturbance background.
2. In the single static camera viewpoint，for the problem of object position offset andloss with the tracking method based on single feature, we present a novel tracking algorithm with a covariance matrix of multiple features and EM-GMM method. Thealgorithm utilizes ME-GMM algorithm to extract moving object region of each frame andremove the background components in the video stream, effectively avoids the targetposition offset in tracking process, and then on the basis of the foreground region'scovariance matrix matching in consecutive frames, we analyze the motion state of movingobjects in the continuous frames to predict the likely positions in the next frameovercoming the problem of missing due to occlusion. In the dataset PETS2009[59],experimental results indicate that, compared with the tracking algorithm based on particlefilter, the proposed tracking algorithm has more accurate tracking position of object andlower loss rate.
3. To the question of accuracy of moving object re-identification with multiplenon-overlapping surveillance cameras, we put forward a performance model of movingobject description based on shape context and AP clustering. Firstly, we utilize APclustering algorithm to determine the clustering centers of the object image in the log colorspace and the category each pixels belonging to, then we put all the clustering centers as areference point set to generate the shape context descriptor set, finally we calculate theEMD distance of objects extracted from different non-overlapping cameras to matching thesame objects, reducing the matching error caused by the illumination change, Angle andscale changes. Experimental results in dataset VIPeR[60]show that the method is effective.
本文针对无重叠视域多摄像机监控环境中不规则扰动背景的运动目标检测问题、固定单摄像机下的目标跟踪问题和多摄像机间的目标匹配问题进行了深入研究，取得了如下成果：
1．针对混合高斯背景模型易将场景中的不规则扰动背景成分当作前景目标区域，造成误检率较高的问题，本文提出了一种融合区域运动估计的改进算法（ME-GMM）。该方法首先利用混合高斯背景模型有效应对场景中规则扰动并对前景目标区域进行预提取，然后通过区域运动估计进一步分析各区域的运动特征来区分不规则扰动背景成分和运动前景目标区域，从而完成实际监控场景的运动目标检测。在数据集I2R[58]实验结果表明，与现有算法相比，本文提出的算法能更好地检测不规则背景扰动情况下的运动目标。
2．针对固定单摄像头视域内基于单一特征的目标跟踪算法容易发生目标位置偏移、目标丢失等问题，本文提出了一种基于多特征协方差矩阵的目标跟踪算法。该算法首先利用ME-GMM运动目标检测算法提取出每一帧的运动目标区域并去除视频流中的背景成分，有效避免了跟踪过程可能发生的目标位置偏移；然后运用多特征协方差矩阵对连续帧间的目标区域进行区域匹配，并依此分析目标在连续帧间的运动状态预测目标在下一帧中可能出现的位置，有效克服了因遮挡导致的跟踪丢失问题。在数据集PETS2009[59]上的实验结果表明，与基于单一特征的粒子滤波的跟踪算法相比，本文提出的目标跟踪算法目标位置更准确，跟踪丢失率更低。
3．针对无重叠视域中目标在不同摄像间外观容易发生变化，使得目标在多摄像机间难以获得准确匹配的问题，在分析颜色不变性特征的基础上，提出了一种基于AP聚类算法的形状上下文目标表现模型。该模型首先通过AP聚类算法确定目标图像在log颜色空间的聚类中心和各像素所属的类别，然后将所有聚类中心作为参照点集合生成形状上下文描述子集合，最后通过计算无重叠视域多摄像机监控环境下获取的不同目标表现模型之间的EMD距离实现相同目标的匹配，降低了光照变化、角度变化和尺度变化所带来的匹配误差。在数据集VIPeR[60]的仿真实验证明，本文提出的目标表现模型可以有效实现无重叠视域多摄像机监控环境下的目标的匹配。
In recent years, with the rapid development of intelligent video surveillance system,as well as the urgent demand of the security situation, people’s requirements for smartintelligent video surveillance system are increasing more and more, and it is muchaccounted of day by day that multi-cameras network because of its expansive surveillanceareas. Nevertheless, on the one hand, due to the consideration of capital and resources, it isunlikely to cover all surveillance area with a great amount of monitors. Therefore, cameratracking with non-overlapping views has become a main part in the study of wide-areavideo surveillance. On the other hand, the precondition in the camera tracking is objectmatching which is the significant research project in the object detection and matching formultiple non-overlapping surveillance cameras.
This thesis contrapose the existing problem in the moving object detection ofdisturbed irregular background, object tracking with a single static camera and objectre-identification for multiple non-overlapping surveillance cameras. With the in-depthstudy, the following results can be achieved:
1. To the question the high rate of false drop of moving target detection with Gaussianmixture model in the irregular disturbed background, combined with the motion estimationmethod usually used in video coding, we proposed an effective method of real-timemoving object detection. Gaussian mixture model can effectively cope with regulardisturbance in the scene, and has an advantage of foreground area extraction, at the sametime, irregular disturbing region in the background will also be detected as movingforeground region. Therefore, with the detection results of Gaussian mixture model, wefurther utilize the regional motion estimation method to differentiate irregular disturbingregion of background and prospect moving object region. The experimental results indataset I2R[58]show that it is very effective that the method detailed in this paper that objectdetection for irregular disturbance background.
2. In the single static camera viewpoint，for the problem of object position offset andloss with the tracking method based on single feature, we present a novel tracking algorithm with a covariance matrix of multiple features and EM-GMM method. Thealgorithm utilizes ME-GMM algorithm to extract moving object region of each frame andremove the background components in the video stream, effectively avoids the targetposition offset in tracking process, and then on the basis of the foreground region'scovariance matrix matching in consecutive frames, we analyze the motion state of movingobjects in the continuous frames to predict the likely positions in the next frameovercoming the problem of missing due to occlusion. In the dataset PETS2009[59],experimental results indicate that, compared with the tracking algorithm based on particlefilter, the proposed tracking algorithm has more accurate tracking position of object andlower loss rate.
3. To the question of accuracy of moving object re-identification with multiplenon-overlapping surveillance cameras, we put forward a performance model of movingobject description based on shape context and AP clustering. Firstly, we utilize APclustering algorithm to determine the clustering centers of the object image in the log colorspace and the category each pixels belonging to, then we put all the clustering centers as areference point set to generate the shape context descriptor set, finally we calculate theEMD distance of objects extracted from different non-overlapping cameras to matching thesame objects, reducing the matching error caused by the illumination change, Angle andscale changes. Experimental results in dataset VIPeR[60]show that the method is effective.
引文
[1] Wang X.G. Intelligent multi-camera video surveillance: A review[J]. PatternRecognition Letters,2013,34(1):3-19.
[2] Collins R.T, Lipton A.J, Kanade T. A system for video surveillance and monitoring[R].Pittsburgh: Carnegie Mellon University,2000:1-68.
[3] Piccardi M. Background subtraction techniques: a review[C]. IEEE InternationalConference on Systems, Man and Cybernetics,2004,4:3099-3104.
[4] Lipton A.J, Fujinyoshi H, Patil R.S. Moving target classification and tracking fromreal-time video[C]. IEEE Workshop on Applications of Computer Vision, Princeton,USA,1998:8-14.
[5] Valera M, Velastin S.A. Intelligent distributed surveillance systems: a review[C]. IEEEInternational Conference on Vision, Image and Signal Processing,2005,152(2):192-204.
[6] Elgammal A, Duraisvami R, Hawood D. Background and foreground modeling usingnonparametric kernel density estimation for visual surveillance[J]. Proceeding of theIEEE,2002,90(7):1151-1163.
[7] Mittal A, Paragios N. Motion-based background subtraction using adaptive kerneldensity estimation[C]. IEEE Computer Society Conference on Computer Vision andPattern Recognition, Washington, DC, USA,2004,2(2):302-309.
[8] Kim K, Harwood D, Davis L.S, et al. Real-time foreground-background segmentationusing codebook model[J]. Real Time Imaging,2005,11(3):172-185.
[9] Benezeth Y, Jodoin P.M, Emile B, Laurent H, Rosenberger C. Review and evaluationof commonly-implemented background subtraction algorithms[C]. IEEE InternationalConference on Pattern Recognition, Tampa, Florida, USA,2008:1-4.
[10]Suchendra M, Bhandarkar, Luo X.Z. Integrated detection and tracking of multiplefaces using particle filtering and optical flow-based elastic matching[J]. Computervision and image understanding,2009,113(6):708-725.
[11]Li Y.B, Chen F, Xu W.L, et al. Gaussian-based codebook model for video backgroundsubtraction[C]. International Conference on Natural Computation, Chongqing, China,2006,9:762-765.
[12]Perez P, Hue C, Vermaak J, Gangnet M. Color-based probabilistic tracking[C].European Conference on Computer Vision, Copenhagen,2002,1:661-675.
[13]Li Y, Ai H, Yamashita T, Lao S, Kawade M. Tracking in low frame rate video: acascade particle filter with discriminative observers of different life spans[J]. IEEETransactions on Pattern Analysis and Machine Intelligence,2008,30(10):1728-1740.
[14]Badenas J, Sanchiz J.M, Pla F, Using temporal integration for tracking regions intraffic monitoring sequences[C], IEEE International Conference on PatternRecognition, Barcelona, Spain,2000,3:1125-1128.
[15]Matthews I, Iahilcawa T, Baker S, The template update problem[J]. IEEE Transactionson Pattern Analysis and Machine Intelligence,2004,26:810-815.
[16]Kaneko T, Hori O. Feature selection for reliable tracking using template matching[C].IEEE Conference on Computer Vision and Pattern Recognition, Madison, WI, USA,2003,1:796-802.
[17]Kass M, Witkin A, Terzopoulos D, Snakes: Active contour models[J]. InternationalJournal of Computer Vision,1988,1(4):321-331.
[18]Osher S, Sethian J.A. Fronts propagating with curvature dependent speed:algorithmsbased on Hamilton-Jacobi formulation[J]. Journal of Computational Physics,1988,79(1):12-49.
[19]Fang H; Kim S-H, Jang J.W. A snake algorithm for automatically tracking multipleobjects[C]. IEEE International Conference on Image Processing, Brussels, Belguim,2011:469-472.
[20]Hu W, Zhou X, Li W, Luo W, Zhang X, Maybank S. Active Contour-Based VisualTracking by Integrating Colors, Shapes, and Motions[J]. IEEE Transactions on ImageProcessing,2013,22(5):1778-1792.
[21]Straw A.D, Branson K, Neumann T.R, Dickinson M.H. Multi-camera realtime3dtracking of multiple flying animals[J], Journal of the Royal Society Interface,2011,8(56):395-409.
[22]Wang X.G, Doretto G, Sebastian T, Rittscher J, Tu P. Shape and appearance contextmodeling[C], IEEE International Conference on Computer Vision, Rio de Janeiro,Brazil,2007:1-8.
[23]Park U, Jain A, Kitahara I, Kogure K, Hagita N, Vise: Visual search engine usingmultiple networked cameras[C], IEEE International Conference on PatternRecognition, Hong Kong, China,2006,3:1204-1207.
[24]Gilbert A, Bowden R. Tracking objects across cameras by incrementally learninginter-camera color calibration and patterns of activity[C]. European Conference onComputer Vision, Hong Kong, China,2006,3952:125-136.
[25]Javed O, Shafique K, Shah M. Appearance Modeling for Tracking in MultipleNon-overlapping Cameras[C]. IEEE Conference on Computer Vision and PatternRecognition, San Diego, CA, USA,2005,2:26-33.
[26]Javed O, Shafique K, Rasheed Z, et al. Modeling Inter-camera Space-time andAppearance Relationships for Tracking Across Non-overlapping Views[J]. ComputerVisual Image Understanding,2007,109(2):146-162.
[27]Mazzeo P.L, Spagnolo P, D’Orazio T. Object Tracking by Non-overlapping DistributedCamera Network[J]. Advanced Concepts for Intelligent Vision Systems,2009,5807:516-527.
[28]Porikli F, Divakaran A. Multi-Camera Calibration, Object Tracking and QueryGeneration[C]. International Conference on Multimedia and Expo, Baltimore,Maryland, USA,2003,1:653-656.
[29]Dalal N, Triggs B. Histograms of oriented gradients for human detection[C]. IEEEConference on Computer Vision and Pattern Recognition, San Diego, CA, USA,2005,1:886-893.
[30]David G.L. Distinctive image features from scale-invariant keypoints[J]. InternationalJournal of Computer Vision,2004,60(2):91-110.
[31]Lo B.P.L, Velastin S.A. Automatic congestion detection system for undergroundplatforms[C]. International Symposium on Intelligent Multimedia, Video and SpeechProcessing, Hong Kong, China,2001:158-161.
[32]Stauffer C, Grimson W.E.L. Learning patterns of activity using real-time tracking[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2000,22(8):747-757.
[33]Zivkovic Z, Heijden F. Efficient adaptive density estimation per image pixel for thetask of background subtraction[J]. Pattern Recognition Letters,2006,27(7):773-780.
[34]Liao S.C, Zhao G.Y, Kellokumpu V, Pietik inen M, Li S.Z. Modeling pixel processwith scale invariant local patterns for background subtraction in complex scenes[C].IEEE Conference on Computer Vision and Pattern Recognition, San Francisco, USA,2010:1301-1306.
[35]Ko T, Soatto S, Estrin D. Warping background subtraction[C]. IEEE Conference onComputer Vision and Pattern Recognition, San Francisco, USA,2010:1331-1338.
[36]Reddy V, Sanderson C, Lovell B.C. Robust foreground object segmentation viaadaptive region-based background modeling[C]. IEEE International Conference onPattern Recognition, Istanbul, Turkey,2010:3939-3942.
[37]Tuzel F.O, Porikli F, Meer P, Region covariance: A fast descriptor for detection andclassification[C]. European Conference on Computer Vision, Graz, Austria,2006,2:589-600.
[38]Bak G.S, Corvee E, Bremond F, Thonnat M. Person re-identification using spatialcovariance regions of human body parts[C], Advanced Video and Signal BasedSurveillance, Boston, USA,2010:435-440.
[39]Heikkila M, Pietikainen M. A texture-based method for modeling the background anddetecting moving objects[J]. IEEE Transactions on Pattern Analysis and MachineIntelligence,2006,28(4):657-662.
[40]Ojala T, Pietikainen M, Maenpaa T. Multiresolution gray-scale and rotation invarianttexture classification with local binary patterns[J]. IEEE Transactions on PatternAnalysis and Machine Intelligence,2002,24(7):971-987.
[41]Xu J, Beaugendre A, Goto S. Real-time human tracking by detection based on HOGand particle filter[C]. International Conference on Computer Sciences andConvergence Information Technology, Jeju Island, South Korea,2011:193-198.
[42]Kviatkovsky I, Adam A, Rivlin E. Color Invariants for Person Re-Identification[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2012, PP(99):1.
[43]Belongie S, Malik J, Puzicha J. Shape matching and object recognition using shapecontexts[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence,2002,24(4):509-522.
[44]孔庆杰.面向广域视频监控的无重叠视域多摄像机目标跟踪[J].2007年全国模式识别学术会议论文集,2007.
[45]Cheng E.D, Madden C, Piccardi M. Mitigating the effects of variable illumination fortracking across disjoint camera views[C]. IEEE International Conference on Video andSignal Based Surveillance, Sydney, NSW, Australia,2006:32-34.
[46]Cheng E.D, Piccardi M. Matching of object moving across disjoint cameras[C]. IEEEInternational Conference on Image Processing, Atlanta, GA, USA,2006:1769-1772.
[47]Cheng E.D, Piccardi M. Disjoint track matching based on a major color spectrumhistogram representation[J]. Optical Engineering,2007(46):1-4.
[48]Swain M, Ballard D. Indexing via color histograms[C]. IEEE International Conferenceon Computer Vision, Kerkyra, Greece,1990:390-393.
[49]Gevers T, Smeulders A. Color-based object recognition[J]. Journal of the PatternRecognition Society,1999,32(3):453-464.
[50]Berwick D, Lee S. A chromaticity space for specularity, illumination color-andillumination pose-invariant3-D object recognition[C]. IEEE International Conferenceon Computer Vision, Bombay, India,1998:165-170.
[51]Finlayson G, Drew M, Funt B. Diagonal transforms suffice for color constancy[C].IEEE International Conference on Computer Vision, Berlin, Germany,1993:164-171.
[52]Finlayson G.D. Spectral sharpening: Sensor transformations for improved colorconstancy[J]. Journal of the Optical Society of America A,1994,11(5):1553-1563.
[53]Rubner Y, Tomasi C, Guibas L. The earth mover’s distance as a metric for imageretrieval[J]. International Journal of Computer Vision,2000,40(2):91-121.
[54]Frey B.J, Dueck D. Clustering by passing messages between data points[J]. Science,2007,315(5814):972-976.
[55]Jeong K, Jaynes C. Object matching in disjoint cameras using a color transferapproach[J]. Machine Vision and Applications,2008,19(5-6):443-455.
[56]Abdel-Hakim A.E, Farag A.A. Csift: a sift descriptor with color invariantcharacteristics[C]. IEEE Computer Society Conference on Computer Vision andPattern Recognition, New York, NY, USA,2006,2:1978-1983.
[57]Moon H, Phillips P. Computational and performance aspects of PCA-based facerecognition algorithms[J]. Perception,2001,30(3):303-3211.
[58]http://perception.i2r.a-star.edu.sg/bk_model/bk_index.html
[59]http://www.cvg.rdg.ac.uk/PETS2009/a.html
[60]http://vision.soe.ucsc.edu/?q=node/178
[2] Collins R.T, Lipton A.J, Kanade T. A system for video surveillance and monitoring[R].Pittsburgh: Carnegie Mellon University,2000:1-68.
[3] Piccardi M. Background subtraction techniques: a review[C]. IEEE InternationalConference on Systems, Man and Cybernetics,2004,4:3099-3104.
[4] Lipton A.J, Fujinyoshi H, Patil R.S. Moving target classification and tracking fromreal-time video[C]. IEEE Workshop on Applications of Computer Vision, Princeton,USA,1998:8-14.
[5] Valera M, Velastin S.A. Intelligent distributed surveillance systems: a review[C]. IEEEInternational Conference on Vision, Image and Signal Processing,2005,152(2):192-204.
[6] Elgammal A, Duraisvami R, Hawood D. Background and foreground modeling usingnonparametric kernel density estimation for visual surveillance[J]. Proceeding of theIEEE,2002,90(7):1151-1163.
[7] Mittal A, Paragios N. Motion-based background subtraction using adaptive kerneldensity estimation[C]. IEEE Computer Society Conference on Computer Vision andPattern Recognition, Washington, DC, USA,2004,2(2):302-309.
[8] Kim K, Harwood D, Davis L.S, et al. Real-time foreground-background segmentationusing codebook model[J]. Real Time Imaging,2005,11(3):172-185.
[9] Benezeth Y, Jodoin P.M, Emile B, Laurent H, Rosenberger C. Review and evaluationof commonly-implemented background subtraction algorithms[C]. IEEE InternationalConference on Pattern Recognition, Tampa, Florida, USA,2008:1-4.
[10]Suchendra M, Bhandarkar, Luo X.Z. Integrated detection and tracking of multiplefaces using particle filtering and optical flow-based elastic matching[J]. Computervision and image understanding,2009,113(6):708-725.
[11]Li Y.B, Chen F, Xu W.L, et al. Gaussian-based codebook model for video backgroundsubtraction[C]. International Conference on Natural Computation, Chongqing, China,2006,9:762-765.
[12]Perez P, Hue C, Vermaak J, Gangnet M. Color-based probabilistic tracking[C].European Conference on Computer Vision, Copenhagen,2002,1:661-675.
[13]Li Y, Ai H, Yamashita T, Lao S, Kawade M. Tracking in low frame rate video: acascade particle filter with discriminative observers of different life spans[J]. IEEETransactions on Pattern Analysis and Machine Intelligence,2008,30(10):1728-1740.
[14]Badenas J, Sanchiz J.M, Pla F, Using temporal integration for tracking regions intraffic monitoring sequences[C], IEEE International Conference on PatternRecognition, Barcelona, Spain,2000,3:1125-1128.
[15]Matthews I, Iahilcawa T, Baker S, The template update problem[J]. IEEE Transactionson Pattern Analysis and Machine Intelligence,2004,26:810-815.
[16]Kaneko T, Hori O. Feature selection for reliable tracking using template matching[C].IEEE Conference on Computer Vision and Pattern Recognition, Madison, WI, USA,2003,1:796-802.
[17]Kass M, Witkin A, Terzopoulos D, Snakes: Active contour models[J]. InternationalJournal of Computer Vision,1988,1(4):321-331.
[18]Osher S, Sethian J.A. Fronts propagating with curvature dependent speed:algorithmsbased on Hamilton-Jacobi formulation[J]. Journal of Computational Physics,1988,79(1):12-49.
[19]Fang H; Kim S-H, Jang J.W. A snake algorithm for automatically tracking multipleobjects[C]. IEEE International Conference on Image Processing, Brussels, Belguim,2011:469-472.
[20]Hu W, Zhou X, Li W, Luo W, Zhang X, Maybank S. Active Contour-Based VisualTracking by Integrating Colors, Shapes, and Motions[J]. IEEE Transactions on ImageProcessing,2013,22(5):1778-1792.
[21]Straw A.D, Branson K, Neumann T.R, Dickinson M.H. Multi-camera realtime3dtracking of multiple flying animals[J], Journal of the Royal Society Interface,2011,8(56):395-409.
[22]Wang X.G, Doretto G, Sebastian T, Rittscher J, Tu P. Shape and appearance contextmodeling[C], IEEE International Conference on Computer Vision, Rio de Janeiro,Brazil,2007:1-8.
[23]Park U, Jain A, Kitahara I, Kogure K, Hagita N, Vise: Visual search engine usingmultiple networked cameras[C], IEEE International Conference on PatternRecognition, Hong Kong, China,2006,3:1204-1207.
[24]Gilbert A, Bowden R. Tracking objects across cameras by incrementally learninginter-camera color calibration and patterns of activity[C]. European Conference onComputer Vision, Hong Kong, China,2006,3952:125-136.
[25]Javed O, Shafique K, Shah M. Appearance Modeling for Tracking in MultipleNon-overlapping Cameras[C]. IEEE Conference on Computer Vision and PatternRecognition, San Diego, CA, USA,2005,2:26-33.
[26]Javed O, Shafique K, Rasheed Z, et al. Modeling Inter-camera Space-time andAppearance Relationships for Tracking Across Non-overlapping Views[J]. ComputerVisual Image Understanding,2007,109(2):146-162.
[27]Mazzeo P.L, Spagnolo P, D’Orazio T. Object Tracking by Non-overlapping DistributedCamera Network[J]. Advanced Concepts for Intelligent Vision Systems,2009,5807:516-527.
[28]Porikli F, Divakaran A. Multi-Camera Calibration, Object Tracking and QueryGeneration[C]. International Conference on Multimedia and Expo, Baltimore,Maryland, USA,2003,1:653-656.
[29]Dalal N, Triggs B. Histograms of oriented gradients for human detection[C]. IEEEConference on Computer Vision and Pattern Recognition, San Diego, CA, USA,2005,1:886-893.
[30]David G.L. Distinctive image features from scale-invariant keypoints[J]. InternationalJournal of Computer Vision,2004,60(2):91-110.
[31]Lo B.P.L, Velastin S.A. Automatic congestion detection system for undergroundplatforms[C]. International Symposium on Intelligent Multimedia, Video and SpeechProcessing, Hong Kong, China,2001:158-161.
[32]Stauffer C, Grimson W.E.L. Learning patterns of activity using real-time tracking[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2000,22(8):747-757.
[33]Zivkovic Z, Heijden F. Efficient adaptive density estimation per image pixel for thetask of background subtraction[J]. Pattern Recognition Letters,2006,27(7):773-780.
[34]Liao S.C, Zhao G.Y, Kellokumpu V, Pietik inen M, Li S.Z. Modeling pixel processwith scale invariant local patterns for background subtraction in complex scenes[C].IEEE Conference on Computer Vision and Pattern Recognition, San Francisco, USA,2010:1301-1306.
[35]Ko T, Soatto S, Estrin D. Warping background subtraction[C]. IEEE Conference onComputer Vision and Pattern Recognition, San Francisco, USA,2010:1331-1338.
[36]Reddy V, Sanderson C, Lovell B.C. Robust foreground object segmentation viaadaptive region-based background modeling[C]. IEEE International Conference onPattern Recognition, Istanbul, Turkey,2010:3939-3942.
[37]Tuzel F.O, Porikli F, Meer P, Region covariance: A fast descriptor for detection andclassification[C]. European Conference on Computer Vision, Graz, Austria,2006,2:589-600.
[38]Bak G.S, Corvee E, Bremond F, Thonnat M. Person re-identification using spatialcovariance regions of human body parts[C], Advanced Video and Signal BasedSurveillance, Boston, USA,2010:435-440.
[39]Heikkila M, Pietikainen M. A texture-based method for modeling the background anddetecting moving objects[J]. IEEE Transactions on Pattern Analysis and MachineIntelligence,2006,28(4):657-662.
[40]Ojala T, Pietikainen M, Maenpaa T. Multiresolution gray-scale and rotation invarianttexture classification with local binary patterns[J]. IEEE Transactions on PatternAnalysis and Machine Intelligence,2002,24(7):971-987.
[41]Xu J, Beaugendre A, Goto S. Real-time human tracking by detection based on HOGand particle filter[C]. International Conference on Computer Sciences andConvergence Information Technology, Jeju Island, South Korea,2011:193-198.
[42]Kviatkovsky I, Adam A, Rivlin E. Color Invariants for Person Re-Identification[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2012, PP(99):1.
[43]Belongie S, Malik J, Puzicha J. Shape matching and object recognition using shapecontexts[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence,2002,24(4):509-522.
[44]孔庆杰.面向广域视频监控的无重叠视域多摄像机目标跟踪[J].2007年全国模式识别学术会议论文集,2007.
[45]Cheng E.D, Madden C, Piccardi M. Mitigating the effects of variable illumination fortracking across disjoint camera views[C]. IEEE International Conference on Video andSignal Based Surveillance, Sydney, NSW, Australia,2006:32-34.
[46]Cheng E.D, Piccardi M. Matching of object moving across disjoint cameras[C]. IEEEInternational Conference on Image Processing, Atlanta, GA, USA,2006:1769-1772.
[47]Cheng E.D, Piccardi M. Disjoint track matching based on a major color spectrumhistogram representation[J]. Optical Engineering,2007(46):1-4.
[48]Swain M, Ballard D. Indexing via color histograms[C]. IEEE International Conferenceon Computer Vision, Kerkyra, Greece,1990:390-393.
[49]Gevers T, Smeulders A. Color-based object recognition[J]. Journal of the PatternRecognition Society,1999,32(3):453-464.
[50]Berwick D, Lee S. A chromaticity space for specularity, illumination color-andillumination pose-invariant3-D object recognition[C]. IEEE International Conferenceon Computer Vision, Bombay, India,1998:165-170.
[51]Finlayson G, Drew M, Funt B. Diagonal transforms suffice for color constancy[C].IEEE International Conference on Computer Vision, Berlin, Germany,1993:164-171.
[52]Finlayson G.D. Spectral sharpening: Sensor transformations for improved colorconstancy[J]. Journal of the Optical Society of America A,1994,11(5):1553-1563.
[53]Rubner Y, Tomasi C, Guibas L. The earth mover’s distance as a metric for imageretrieval[J]. International Journal of Computer Vision,2000,40(2):91-121.
[54]Frey B.J, Dueck D. Clustering by passing messages between data points[J]. Science,2007,315(5814):972-976.
[55]Jeong K, Jaynes C. Object matching in disjoint cameras using a color transferapproach[J]. Machine Vision and Applications,2008,19(5-6):443-455.
[56]Abdel-Hakim A.E, Farag A.A. Csift: a sift descriptor with color invariantcharacteristics[C]. IEEE Computer Society Conference on Computer Vision andPattern Recognition, New York, NY, USA,2006,2:1978-1983.
[57]Moon H, Phillips P. Computational and performance aspects of PCA-based facerecognition algorithms[J]. Perception,2001,30(3):303-3211.
[58]http://perception.i2r.a-star.edu.sg/bk_model/bk_index.html
[59]http://www.cvg.rdg.ac.uk/PETS2009/a.html
[60]http://vision.soe.ucsc.edu/?q=node/178