基于图像序列的行人检测与跟踪算法的研究
|
摘要
智能楼宇中的视频监控已经受到社会越来越多的关注。智能楼宇中的视频监控是利用数字图像处理、机器学习以及计算机视觉等技术,自动且实时的处理和分析视频图像,对视频监控场景中存在的可疑行人目标、潜在危险或违规行为进行预警或报警。
本文针对光照变化,遮挡,快速运动等场景下的行人检测和跟踪问题,深入研究了基于分类器的的行人检测和跟踪方法,本文主要研究工作如下：
1)针对光照变化,遮挡,快速运动等场景下的行人检测问题,提出了一种针对感兴趣区域,基于梯度方向直方图和纹理特征相融合的行人检测方法。该方法通过行人跟踪片在行人检测过程中引入时间相关性,即在行人检测过程中引入行人跟踪结果以调整行人检测器中的扫描尺度,缩小检测器扫描的层次和范围。使得行人检测器只针对感兴趣区域(存在行人概率高的区域)进行检测,提高了行人识别率。分类器的特征使用改进的梯度方向直方图和纹理相融合特征。从实验结果发现这种融合的特征不仅提高了分类器的性能,使得行人识别效果上有所提高,而且在一定程度上能够解决行人的局部遮挡问题。
2)针对行人跟踪“漂移”问题,研究和分析了基于分类器的跟踪算法。本文提出了将离线行人检测和在线行人跟踪相结合的行人跟踪方法。将离线行人检测作为行人的先验知识引入到跟踪器中,它不依赖自学习过程来更新,因此避免自学习造成错误累积而产生“漂移”。该方法不仅解决了在线行人跟踪的初始化问题,也解决了在线分类器的训练数据的选择问题。由于引入了离线行人分类,当行人被遮挡后重新出现在场景或者跟踪器跟丢的情况下,该方法能重新识别行人,继续对其跟踪,因而实现了长时间行人跟踪。
More and more people are paying attention to intelligent video surveillance. Intelligent video surveillance uses technologies such as digital image processing, artificial intelligence and computer vision to process and analyze videos automatically in real-time, which can warn or alarm suspicious targets, potential hazards or violations in monitored scenes.
This dissertation researched pedestrian detection and tracking algorithm under different conditions, including illumination change, occlusion and fast movement.Main work of this dissertationis listed as follows:
1) To solve the problem ofpedestrian detection and tracking under different conditions, including illumination change, occlusion and fast movement, this dissertation proposed a pedestrian detection algorithm fusing histogram of gradient direction and textural features aiming at the interested areas. During pedestrian detection process, detector adjusts the scanning scale of the pedestrian detector by introducing pedestrian tracking result information to reduce the level and scope of the scanning of the detector. The detector only detects in the region of interest (The area where there is a high probability of pedestrian) to improve the pedestrian detection rate.The features merged histogram of gradient merging and textural features. Experimental results show that the improved histograms of gradient merging with local binary pattern scan not only improve the recognition rate, but can alsoand solve the occlusion problem to a certain extent.
2)To solve the "drift" problem in pedestrians tracking, this dissertation researched and analyzed pedestrian tracking algorithm based on classifier.We proposed a method to combine offline pedestrian trackingand online pedestrian tracking templates. The Offline pedestrian detection results can introduce priori knowledgeinto the tracker. And it did not depend on the learning process to update so that it can avoid "drift" caused by the error accumulation of learning. This algorithm not only solved the problem of online pedestrian tracking initialization, but also solved the problem of selection of training data. As offline pedestrian classification, when pedestrians were presented in the scenes again after being occluded or the tracker lost targets, this method can re-recognize the pedestrians and keep tracking, so that the long time tracking can be achieved.
本文针对光照变化,遮挡,快速运动等场景下的行人检测和跟踪问题,深入研究了基于分类器的的行人检测和跟踪方法,本文主要研究工作如下：
1)针对光照变化,遮挡,快速运动等场景下的行人检测问题,提出了一种针对感兴趣区域,基于梯度方向直方图和纹理特征相融合的行人检测方法。该方法通过行人跟踪片在行人检测过程中引入时间相关性,即在行人检测过程中引入行人跟踪结果以调整行人检测器中的扫描尺度,缩小检测器扫描的层次和范围。使得行人检测器只针对感兴趣区域(存在行人概率高的区域)进行检测,提高了行人识别率。分类器的特征使用改进的梯度方向直方图和纹理相融合特征。从实验结果发现这种融合的特征不仅提高了分类器的性能,使得行人识别效果上有所提高,而且在一定程度上能够解决行人的局部遮挡问题。
2)针对行人跟踪“漂移”问题,研究和分析了基于分类器的跟踪算法。本文提出了将离线行人检测和在线行人跟踪相结合的行人跟踪方法。将离线行人检测作为行人的先验知识引入到跟踪器中,它不依赖自学习过程来更新,因此避免自学习造成错误累积而产生“漂移”。该方法不仅解决了在线行人跟踪的初始化问题,也解决了在线分类器的训练数据的选择问题。由于引入了离线行人分类,当行人被遮挡后重新出现在场景或者跟踪器跟丢的情况下,该方法能重新识别行人,继续对其跟踪,因而实现了长时间行人跟踪。
More and more people are paying attention to intelligent video surveillance. Intelligent video surveillance uses technologies such as digital image processing, artificial intelligence and computer vision to process and analyze videos automatically in real-time, which can warn or alarm suspicious targets, potential hazards or violations in monitored scenes.
This dissertation researched pedestrian detection and tracking algorithm under different conditions, including illumination change, occlusion and fast movement.Main work of this dissertationis listed as follows:
1) To solve the problem ofpedestrian detection and tracking under different conditions, including illumination change, occlusion and fast movement, this dissertation proposed a pedestrian detection algorithm fusing histogram of gradient direction and textural features aiming at the interested areas. During pedestrian detection process, detector adjusts the scanning scale of the pedestrian detector by introducing pedestrian tracking result information to reduce the level and scope of the scanning of the detector. The detector only detects in the region of interest (The area where there is a high probability of pedestrian) to improve the pedestrian detection rate.The features merged histogram of gradient merging and textural features. Experimental results show that the improved histograms of gradient merging with local binary pattern scan not only improve the recognition rate, but can alsoand solve the occlusion problem to a certain extent.
2)To solve the "drift" problem in pedestrians tracking, this dissertation researched and analyzed pedestrian tracking algorithm based on classifier.We proposed a method to combine offline pedestrian trackingand online pedestrian tracking templates. The Offline pedestrian detection results can introduce priori knowledgeinto the tracker. And it did not depend on the learning process to update so that it can avoid "drift" caused by the error accumulation of learning. This algorithm not only solved the problem of online pedestrian tracking initialization, but also solved the problem of selection of training data. As offline pedestrian classification, when pedestrians were presented in the scenes again after being occluded or the tracker lost targets, this method can re-recognize the pedestrians and keep tracking, so that the long time tracking can be achieved.
引文
[1].焦波,面向智能视频监控的运动目标检测与跟踪方法研究[D]:[博士].长沙：国防科学技术大学,4-7,2009.
[2].万成凯.无标记人体运动捕捉及姿态估计的研究[学位论文].北京：北京交通大学,2009.
[3].赵克栋.视频监控系统设计与工程应用[D].北京邮电大学2009
[4].张可义,岳秀江,韩立新.视频监控系统工程应用研究[J].制造业自动化2007年03期.
[5].Christopher Richard Wren, Ali Azarbayejani, Trevor Darrell, and Alex Paul Pentland. "Pfinder:real-time tracking of the human body"[j]. IEEE Trans Pattern Analysis and Machine Intelligence,1997,19(7).
[6].Zhao L, Thorpe C. Stereo and neural network- based pedestrian detection[J] IEEE Transactions on Intelligent Transportation Systems,2000,1(3):148-154.
[7].Haritaoglu, harwood D and Davis L. W4:real-time surveillance of people and their activities. IEEE Trans Pattern Analysis and Machine Intelligence, 2000,22(8):809-830.
[8].Cao X B, Qiao H, Wang F Y, Zhang X Z. Application of cooperative co-evolution in pedestrian detection systems[A]. Proceeding s of IEEE International Conference on Intelligence and Security Informatics [C]. Atlanta, USA:Springer,2005.3495.664-665.
[9].Wei C X, Cao X B, Y W Xu, Qiao H,Wang F Y. The treelike assembly classifier for pedestrian detection[A].Proceedings of Pacific Asian Workshop on Intelligence and Security Informatics[C]. Chengdu, China:Springer,2007. 232-237.
[10].贾慧星,章毓晋.车辆辅助驾驶系统中基于计算机视觉的行人检测研究综述[J].自动化学报,2007,33(1)：84-90.
[11].Tian Q M, Sun H, Luo Y P, Hu D C. Nighttime pedestrian detection with a normal camera using SVM classifier[A]. Proceedings of International Symposium on Neural Networks[C].Chongqing, China:Springer,2005.3497. 189-194.
[12].张浩.视频运动人体行为识别与分类方法研究[学位论文].西安：西安电子科技大学,2011.
[13]. D. M. Gavrila and S. Munder. Multi-cue pedestrian detection and tracking from a moving vehicle. IJCV, pages 41-59,2007.
[14]. LinZhe, L S Davis, D Doermannet, al. Hierarchical part-temple matching for human detection and degmentation. Proceedings of IEEE International Conference on Computer Vision. Rio de Janeiro, Brasil:IEEE Press,2007:1-8.
[15]. B. Wu and R. Nevatia, Detection and tracking of multiple, partially occluded humans by bayesian combination of edgelet based part detectors,IJCV, Vol.75, No.2, pp.247-266,2007.
[16]. P. Felzenszwalb, D. McAllester, and D. Ramanan. A discriminatively trained, multiscale, deformable part model. In CVPR,2008.
[17].C.Papageorgiou and T.Poggio, A trainable system for object detection. IJCV, Vol.38, No.1, pp.15-33,2000.
[18].N.Dalal and B.Triggs, Histograms of oriented gradients for human detection, n CVPR, Vol, pages 886-893,2005.
[19]. Y. Mu, S. Yan, Y. Liu, T. Huang, and B. Zhou, Discriminative local binary patterns for human detection in personal album,in CVPR,2008.
[20]. P.A. Viola and M.J. Jones, Robust real-time face detection.IJCV, Vol.57, No. 2, pp.137-154,2004.
[21]. X. Wang, T. X. Han, and S. Yan, An HOG-LBP human detector with partial occlusion handling, in ICCV,2009.
[22]. Levi K, Y Weiss. Learning object detection from a small number of examples: the importance of a good features. Proceedings of IEEE Computer Society Conference on Computer Vision and Pattern Recognition. Washington, DC, USA:IEEE Computer Society Press,2:53-60,2004.
[23]. P. Sabzmeydani and G. Mori. Detecting pedestrians by learningshapelet features. In CVPR,2007.
[24]. N.Dalal, Finding people in images and videos. PhD thesis, INRIA Rhne-Alpes, Grenoble, France,2006.
[25]. S. Avidan, Ensemble tracking. PAMI,29:261-271,2007.
[26]. Grabner, H, Bischof, H, On-line boosting and vision, ProceedingsConference on Computer Vision and Pattern Recognition.Vol.1,IEEE Computer Society, New York, NY, pp.260-267,2006.
[27].边肇祺,张学工.模式识别(第二版),清华大学出版社,2000.
[28]. D. G. Lowe. Distinctive image features from scale-invariant keypoints. International Journal of Computer Vision,60:91-110,2004.
[29]. T. Ojala, M. Pietik"ainen, and D. Harwood. A comparative study of texture measures with classification based on featured distributions. Pattern Recognition,29(1):51-59,1996.
[30].Zdenek Kalal, Krystian Mikolajczyk, and Jiri Matas, Tracking-learning-detection. PAMI, Vol.6, No.1 2010.
[31]. Viola.P, Jones.M, Rapid object detection using a boosted cascade of simple features. Proceedings Conference on Computer Vision and Pattern Recognition. Vol.1.IEEE Computer Society, Kauai, HI, pp.511-518.2001.
[32]. J. Friedman, T. Hasite, and R. Tibshirani, "Additive Logistic Regression:A Statistical View of Boosting," technical report, Dept.of Statistics, Stanford Univ.,1998.
[33]. Wu Bo, Ai Haizhou, Huang Chang. LUT-based Adaboost for gender classification [A]. In:Proceedings of International Conference on Audio-and Video-Based Biometric Person Authentication[C]. Guildford:2003.104-110.
[34]. Y. Freund and R.E. Schapire, "A Decision-Theoretic Generalization of On-Line Learning and an Application to Boosting," Proc. Second European Conf. Computational Learning Theory, pp.23-37,1995.
[35]. Oza. N, Russell. S, Online bagging and boosting. Proceedings Artificial Intelligence and Statistics. Morgan Kaufmann, Florida, pp.105-112.2001.
[36]. Robust Object Tracking with Online Multiple Instance Learning, in CVPR, 2009.
[37].Yizong Cheng. Mean Shift, Mode Seeking and Clustering [J]. IEEEE Transactions On Pattern Analysis And Machaine Intelligence,1995,17(8): 790-799.
[2].万成凯.无标记人体运动捕捉及姿态估计的研究[学位论文].北京：北京交通大学,2009.
[3].赵克栋.视频监控系统设计与工程应用[D].北京邮电大学2009
[4].张可义,岳秀江,韩立新.视频监控系统工程应用研究[J].制造业自动化2007年03期.
[5].Christopher Richard Wren, Ali Azarbayejani, Trevor Darrell, and Alex Paul Pentland. "Pfinder:real-time tracking of the human body"[j]. IEEE Trans Pattern Analysis and Machine Intelligence,1997,19(7).
[6].Zhao L, Thorpe C. Stereo and neural network- based pedestrian detection[J] IEEE Transactions on Intelligent Transportation Systems,2000,1(3):148-154.
[7].Haritaoglu, harwood D and Davis L. W4:real-time surveillance of people and their activities. IEEE Trans Pattern Analysis and Machine Intelligence, 2000,22(8):809-830.
[8].Cao X B, Qiao H, Wang F Y, Zhang X Z. Application of cooperative co-evolution in pedestrian detection systems[A]. Proceeding s of IEEE International Conference on Intelligence and Security Informatics [C]. Atlanta, USA:Springer,2005.3495.664-665.
[9].Wei C X, Cao X B, Y W Xu, Qiao H,Wang F Y. The treelike assembly classifier for pedestrian detection[A].Proceedings of Pacific Asian Workshop on Intelligence and Security Informatics[C]. Chengdu, China:Springer,2007. 232-237.
[10].贾慧星,章毓晋.车辆辅助驾驶系统中基于计算机视觉的行人检测研究综述[J].自动化学报,2007,33(1)：84-90.
[11].Tian Q M, Sun H, Luo Y P, Hu D C. Nighttime pedestrian detection with a normal camera using SVM classifier[A]. Proceedings of International Symposium on Neural Networks[C].Chongqing, China:Springer,2005.3497. 189-194.
[12].张浩.视频运动人体行为识别与分类方法研究[学位论文].西安：西安电子科技大学,2011.
[13]. D. M. Gavrila and S. Munder. Multi-cue pedestrian detection and tracking from a moving vehicle. IJCV, pages 41-59,2007.
[14]. LinZhe, L S Davis, D Doermannet, al. Hierarchical part-temple matching for human detection and degmentation. Proceedings of IEEE International Conference on Computer Vision. Rio de Janeiro, Brasil:IEEE Press,2007:1-8.
[15]. B. Wu and R. Nevatia, Detection and tracking of multiple, partially occluded humans by bayesian combination of edgelet based part detectors,IJCV, Vol.75, No.2, pp.247-266,2007.
[16]. P. Felzenszwalb, D. McAllester, and D. Ramanan. A discriminatively trained, multiscale, deformable part model. In CVPR,2008.
[17].C.Papageorgiou and T.Poggio, A trainable system for object detection. IJCV, Vol.38, No.1, pp.15-33,2000.
[18].N.Dalal and B.Triggs, Histograms of oriented gradients for human detection, n CVPR, Vol, pages 886-893,2005.
[19]. Y. Mu, S. Yan, Y. Liu, T. Huang, and B. Zhou, Discriminative local binary patterns for human detection in personal album,in CVPR,2008.
[20]. P.A. Viola and M.J. Jones, Robust real-time face detection.IJCV, Vol.57, No. 2, pp.137-154,2004.
[21]. X. Wang, T. X. Han, and S. Yan, An HOG-LBP human detector with partial occlusion handling, in ICCV,2009.
[22]. Levi K, Y Weiss. Learning object detection from a small number of examples: the importance of a good features. Proceedings of IEEE Computer Society Conference on Computer Vision and Pattern Recognition. Washington, DC, USA:IEEE Computer Society Press,2:53-60,2004.
[23]. P. Sabzmeydani and G. Mori. Detecting pedestrians by learningshapelet features. In CVPR,2007.
[24]. N.Dalal, Finding people in images and videos. PhD thesis, INRIA Rhne-Alpes, Grenoble, France,2006.
[25]. S. Avidan, Ensemble tracking. PAMI,29:261-271,2007.
[26]. Grabner, H, Bischof, H, On-line boosting and vision, ProceedingsConference on Computer Vision and Pattern Recognition.Vol.1,IEEE Computer Society, New York, NY, pp.260-267,2006.
[27].边肇祺,张学工.模式识别(第二版),清华大学出版社,2000.
[28]. D. G. Lowe. Distinctive image features from scale-invariant keypoints. International Journal of Computer Vision,60:91-110,2004.
[29]. T. Ojala, M. Pietik"ainen, and D. Harwood. A comparative study of texture measures with classification based on featured distributions. Pattern Recognition,29(1):51-59,1996.
[30].Zdenek Kalal, Krystian Mikolajczyk, and Jiri Matas, Tracking-learning-detection. PAMI, Vol.6, No.1 2010.
[31]. Viola.P, Jones.M, Rapid object detection using a boosted cascade of simple features. Proceedings Conference on Computer Vision and Pattern Recognition. Vol.1.IEEE Computer Society, Kauai, HI, pp.511-518.2001.
[32]. J. Friedman, T. Hasite, and R. Tibshirani, "Additive Logistic Regression:A Statistical View of Boosting," technical report, Dept.of Statistics, Stanford Univ.,1998.
[33]. Wu Bo, Ai Haizhou, Huang Chang. LUT-based Adaboost for gender classification [A]. In:Proceedings of International Conference on Audio-and Video-Based Biometric Person Authentication[C]. Guildford:2003.104-110.
[34]. Y. Freund and R.E. Schapire, "A Decision-Theoretic Generalization of On-Line Learning and an Application to Boosting," Proc. Second European Conf. Computational Learning Theory, pp.23-37,1995.
[35]. Oza. N, Russell. S, Online bagging and boosting. Proceedings Artificial Intelligence and Statistics. Morgan Kaufmann, Florida, pp.105-112.2001.
[36]. Robust Object Tracking with Online Multiple Instance Learning, in CVPR, 2009.
[37].Yizong Cheng. Mean Shift, Mode Seeking and Clustering [J]. IEEEE Transactions On Pattern Analysis And Machaine Intelligence,1995,17(8): 790-799.