Surveillance Video Anomaly Detection Via Hidden Markov Model

In recent years,terrorist attack happened frequently in the whole world,which makes people paying more attention to the safety of public places.However,the traditional manual detection means can not efficiently and accurately detect abnormal events,so people want computers to automatically analyze images captured by surveillance video,and timely alarm when abnormal events,set up to realize intelligent monitoring system.As a result,in these days when computer vision technology is getting more popular,people want the monitor system can analysis the screen fames automatically,and show alert when abnormal events happen,so as to achieve the construction of intelligent monitoring system..The direction of the research in this field is that anomaly events in the screen are detected by a semi-supervised model.The mean problems are the model compatibility and high computational complexity.In view of this problems,this paper proposes a model to establish a normal event using the hidden Markov model,and the compatibility of the model is enhanced.At the same time,the basic and fast algorithm is used to improve the overall efficiency of the model in feature extraction and selection.In this paper,the hidden Markov model is used to detect abnormal events of the monitoring vides as the basic model.Hidden Markov model is a novel generation model in the anomaly detection field,which can effectively detect different kinds of abnormal events in different scenarios,including position anomalies,velocity anomalies,directional anomalies,and group behavior anomalies.The algorithmic framework of this paper is as follows: First,three-dimensional gradient features are extracted;Features are then randomly selected to improve the efficiency of the model under required accuracy;Next,we use the principal component analysis method to extract the main information and solve the redundant information.In the training process,we establish the normal event model,while for the test data,we calculate the probability of how this data can be generated from this model,and anomalies are determined by the threshold.In this paper,our algorithmic successfully detects abnormal events on three public database.The comparison experiments are carried out to compare with the results of other classical excellent algorithms to obtain the ROC curve and the result in the public recognition of the evaluation mechanism is desirable.