期刊名称：International Journal of Mechatronics, Electrical and Computer Technology
印刷版ISSN：2305-0543
出版年度：2014
卷号：4
期号：12
页码：1249-1273
出版社：Austrian E-Journals of Universal Scientific Organization
摘要：In this paper, an unsupervised framework is presented to learn motion patterns by using hierarchical Bayesian models. It is also employed for activity analysis in visual surveillance. In this research, the concept of activities as motion patterns is considered as a correspondence to far-field camera view. Objects are tracked by using low-level features and then the location and speed of objects are computed as a feature along with trajectories. Under LDA probabilistic model, activities' distributions are learned in feature space. Since there is not an analytic solution for these models, variational inference method is used to approximate latent parameters of the model. This approach is separately measured on the captured data of several cameras and acceptable results are obtained.
