An Online HDP Mixture Model for Video Mining

doi:10.23940/ijpe.18.11.p32.28642876

Int J Performability Eng ›› 2018, Vol. 14 ›› Issue (11): 2864-2876.doi: 10.23940/ijpe.18.11.p32.28642876

Previous Articles Next Articles

An Online HDP Mixture Model for Video Mining

Lin Tang^a, Lin Liu^b, Mingjing Tang^c, and Yu Sun^{b, *}

^a Key Laboratory of Educational Informatization for Nationalities Ministry of Education, Yunnan Normal University, Kunming, 650500, China;
^b School of Information, Yunnan Normal University, Kunming, 650500, China;
^c President Office, Yunnan Normal University, Kunming, 650500, China

Submitted on ;
Contact: * E-mail address: speed8101@sina.com
About author:Lin Tang graduated from Shanghai Jiaotong University with a bachelor's degree and Chinese Academy of Armament Sciences with a Ph.D. He worked as an engineer at China North Industries Group Corporation Limited from 2004 to 2015. He is currently a senior engineer at Yunnan Normal University. His main research interests include video mining and probabilistic graphical models.Lin Liu graduated from the School of Information at Yunnan University with a bachelor's degree, Master's degree, and Ph.D. She is currently a lecturer at Yunnan Normal University. Her main research interests include bioinformatics and machine learning.Mingjing Tang is a Ph.D. candidate at Yunnan University and a lecturer at Yunnan Normal University. His research focuses on software engineering, process mining, and machine learning.Yu Sun received her bachelor's degree from East China Normal University, Master's degree from Yunnan Normal University, and Ph.D. from Chinese Academy of Sciences. She is currently a professor at Yunnan Normal University. Her current research interests include intelligent education.

Abstract

Abstract: In this paper, we address two problems in video mining: real-time inference and the automatic decision of the number of activities in videos. To solve these problems, we present a real-time Bayesian non-parametric model that is able to discover activities and interactions of videos in real-time. In this model, there are two layers modeled in each scene, which are activities and interactions. An activity is represented as the distribution over visual words, and an interaction is represented as the distribution over activities. Then, the Hierarchical Dirichlet Process (HDP) model connects these two layers of video and automatically decides the number of clusters. Moreover, we developed a hybrid stochastic variational Gibbs sampling algorithm for inferring the parameters of the HDP mixture model. This online inference algorithm has the capacity to process the massive video stream dataset. Finally, the detailed experimental results in a crowded traffic scene and a simulated dataset are described and reveal that our online HDP mixture model achieves superior performance in real-time anomaly activity detection.

Key words: video mining, non-parametric model, inference algorithm

Lin Tang, Lin Liu, Mingjing Tang, and Yu Sun. An Online HDP Mixture Model for Video Mining [J]. Int J Performability Eng, 2018, 14(11): 2864-2876.

Add to citation manager EndNote|Reference Manager|ProCite|BibTeX|RefWorks

References

[1] V. Kaltsa, A. Briassouli,I. Kompatsiaris, “Multiple Hierarchical Dirichlet Processes for Anomaly Detection in Traffic,” Computer Vision and Image Understanding, Vol. 169, pp. 28-39, 2018
[2] L. Tang, L. Liu,J. Su, “Real-Time Video Mining based on SNGRLD-rLDA Model,” inProceedings of IEEE International Conference on Intelligent Human-Machine Systems & Cybernetics, pp. 159-164, 2014
[3] D. M. Blei, “Probabilistic Topic Models,” Communications of the ACM, Vol. 55, No. 4, pp. 77-84, 2012
[4] M. D. Hoffman, D. M. Blei,F. Bach, “Online Learning for Latent Dirichlet Allocation,” inProceedings of International Conference on Neural Information Processing Systems, pp. 856-864, Curran Associates Inc, 2010
[5] Y. W. Teh, M. I. Jordan,M. J. Beal, “Hierarchical Dirichlet Processes,” Publications of the American Statistical Association, Vol. 101, No. 476, pp. 1566-1581, 2006
[6] D. Larlus, J. Verbeek,F. Jurie, “Category Level Object Segmentation by Combining Bag-of-Words Models with Dirichlet Processes and Random Fields,” International Journal of Computer Vision, Vol. 88, No. 2, pp. 238-253, 2010
[7] X. Wang, X. Ma,W. E. L.Grimson, “Unsupervised Activity Perception in Crowded and Complicated Scenes Using Hierarchical Bayesian Models,” IEEE Transactions on Pattern Analysis & Machine Intelligence, Vol. 31, No. 3, pp. 539-555, 2009
[8] O. Isupova, D. Kuzin,L. Mihaylova, “AnomalyDetection in Video with Bayesian Nonparametrics,” inProceedings of ICML2016 Anomaly Detection Workshop, 2016
[9] D. Kuettel, M. D. Breitenstein,L. V. Gool, “What’s Going on? Discovering Spatio-Temporal Dependencies in Dynamic Scenes,” inProceedings of 2010 IEEE Computer Society Conference on Computer Vision and Pattern Recognition, pp. 1951-1958, IEEE, 2010
[10] M. Y. Yang, W. Liao, Y. Cao,B. Rosenhahn, “Video Event Recognition and Anomaly Detection by Combining Gaussian Process and Hierarchical Dirichlet Process Models,” Photogrammetric Engineering & Remote Sensing, Vol. 84, No. 4, 2018
[11] M. Gasparini, “Markov Chain Monte Carlo in Practice,” Technometrics, Vol. 39, No. 3, pp. 338-338, 1999
[12] B. Schölkopf, J. Platt,T. Hofmann, “A Collapsed Variational Bayesian Inference Algorithm for Latent Dirichlet Allocation,”Advances in Neural Information Processing Systems, Vol. 19, 2007
[13] L. Tang, L. Liu,J. H. Gan, “A Regional Topic Model Using Hybrid Stochastic Variational Gibbs Sampling for Real-Time Video Mining,” Algorithms, Vol. 11, No. 7, pp. 97, 2018
[14] D. Mimno, M. D. Hoffman,D. M. Blei, “Sparse Stochastic Inference for Latent Dirichlet Allocation,” Vol. 3, pp. 362-365, 2012
[15] M. Girolami and B. Calderhead, “Riemann Manifold Langevin and Hamiltonian Monte Carlo Methods,” Journal of the Royal Statistical Society, Vol. 73, No. 2, pp. 123-214, 2015
[16] M. Welling and Y. W. Teh, “Bayesian Learning via Stochastic Gradient Langevin Dynamics,” inProceedings of International Conference on International Conference on Machine Learning, pp. 681-688, Omnipress, 2011
[17] S. Patterson and Y. W. Teh, “Stochastic Gradient Riemannian Langevin Dynamics on the Probability Simplex,”Advances in Neural Information Processing Systems, pp. 3102-3110, 2013
[18] O. Isupova, D. Kuzin,L. Mihaylova, “Learning Methods for Dynamic Topic Modeling in Automated Behavior Analysis,”IEEE Transactions on Neural Networks and Learning Systems, Vol. 99, pp. 1-14, 2018
[19] Junction Dataset, (http://www.eecs.qmul.ac.uk/~sgg/QMUL_Junction_Datasets/Junction/Junction.html, last accessed on March 5 2017

An Online HDP Mixture Model for Video Mining

PDF

Knowledge

Abstract

Cite this article

share this article

References

Related Articles 0

Recommended 0