Research On Crowd Abnormal Detection Based On Sparse Linear Model And IHMM

With the rapid development of social economy,many developed areas of population movements also increased,many public places such as schools,shopping malls,stations and other scenes that high-density population gathered often lead to a number of dangerous and violent incidents.Therefore,the security problem of public places has a growing demand for intelligent monitoring.As a core subject of intelligent monitoring,video anomaly detection is of great significance to the maintenance of social public order.Based on anomaly detection,this paper studies the crowd feature-sensing and anomaly detection model,and expounds the principle of the algorithm.The details are as follows:In the video scene,since the group is composed of individuals and each individual can show two characteristics: consistent group and different randomness,the goal of group feature perception is to extract the consistency characteristics of each individual,and it is quite difficult to extract the individual characteristics of the individual directly.Considering the sparsity of sparse vectors,this paper proposes crowd feature-sensing algorithm based on Sparse Linear Model,which is a statistical model of probability.It is assumed that the group features have a certain super-Gaussian prior distribution,and Sparse Bayesian Learning and variational inference are used to obtain the sparse representation of the group feature,so as to effectively separate the group feature from the randomness,and extract the consistency characteristics of the individual.Finally,the sparse representation of the image is simulated on the MATLAB platform,and the feasibility and validity of the super-Gaussian prior distribution conjecture are verified.Since anomalous events occur not only in the time domain,but also even occur in the spatial-temporal domain,considering the complexity of anomaly events,based on the sparse linear model and the iHMM statistical model,an anomaly detection method based on SLM-iHMM is proposed,which is based on SLM-i HMM,which is a directed cascade algorithm that extracts and integrates image features in both spatial and temporal domains.In the spatial domain,the statistical representation of group feature and randomness of the individuals are established based on SLM.The group feature is represented in the spatial domain.In the temporal domain,feature is input as iHMM statistic model and the temporal relation of the feature is captured among the transitions of states.In the crowd detection,the SLM-iHMM model of normal events is trained and the test video sequence is as the input of the trained spatial-temporal model.And the logarithm likelihood function is constructed by comparing the training samples and the test samples through the modeltraining parameters for the judgment of abnormal frame by quantitative.Experiments on two open datasets of UMN and PETS2009 show that SLM-iHMM model has remarkable performance in anomaly detection.