期刊名称：Journal of Theoretical and Applied Information Technology
印刷版ISSN：1992-8645
电子版ISSN：1817-3195
出版年度：2018
卷号：96
期号：11
出版社：Journal of Theoretical and Applied
摘要：The concept of crowd refers to the gathering many people in one place, such as train stations, airports and subways, as well as gatherings sports, religious special crowds Hajj and Umrah became highly congested. In this paper, a novel approach is investigated to the crowd behaviors of individual using discriminative models. The novelty of the proposed approach can be described in three aspects. First, we sectioned video segments into spatio-temporal flow-blocks which allow the marginalization of arbitrarily dense flow field. Second, the observed flow field in each flow-block is treated as 2D distribution of samples and mixtures of Gaussian is used to parameterize it by keeping the generality of flow field intact. Moreover, we implemented, K-means algorithm to initialize the mixture model while Expectation Maximization algorithm is employed for optimization. These mixtures of Gaussian result in the distinct flow patterns (i.e. precisely a sequence of dynamic patterns) for each flow-block. Third, discriminative models such as Conditional Random Field, Hidden Conditional Random Field and Latent-dynamic Conditional Random Field were employed one for each flow-block and were learned from the sequence of dynamic patterns which was then classified for each flow-block as normal and abnormal. Our experiment on our own realistic Data set (Hajj-Umrah DataSet) from crowd in the pilgrimage shows promising results with no scarifying real-time performance for a wide range of practical crowd applications.
关键词：Crowd Behavior; Gaussian mixture; K-means technique; conditional Random Field.
