Abnormal event detection based on social force model combined with crowd violent flow

In this study, we use the social force model (SFM) combined with the violent flows descriptor (vif) to detect abnormal events. Our method extracts the local density to calculate the parameters of SFM and the vif to represent the crowd stability status. The Latent Dirichlet Allocation(LDA) model is established based on the bag-of-words model combined with the temporal and spatial features of visual words, and then the maximum likelihood function is used to identify abnormal events.