Sudden The Crowd Incident Intelligent Video Surveillance

With the development of economy and society, especially the progress of urbanization in our country, the population density of cities increases continuously.The public traffic and infrastructure in cities often meet crowd flow with great density in rush hour,which may result in accidents if the population congestion call not be evacuated in time. In addition, it could easily lead to some unexpected crowd incidents because of the increasing of population density,which also threatens urban security. Therefore, it's very urgent to research how to monitor these incidents effectively.A method of monitoring unexpected crowd incidents is applied by this dissertation. First, it's necessary to get the feature of density of crowd in monitoring scene. And then it gets the blocking rate of crowd in monitoring scene. At last these features are sent to classifier. And the classifier will determine whether there has been an unexpected crowd incident.The development and basic principles of crowd monitoring system both in our country and abroad is introduced by this dissertation.The simplest way of crowd density estimation is pixel-based crowd estimation. But it's applied only to the low-density scene. For the high-density scene, there is usually used analysis method based on texture estimation. However, texture estimation is more complex than pixel-based crowd estimation.The density of crowed is often high in the unexpected crowd incidents. So this paper decided to adopt texture estimation to extract the characteristics of population density. In the feature extraction process, we will use Gray Level Dependence Matrix and its statistical characteristic values, such as Entropy, Energy and Contrast.However, it can not detect the unexpected crowd incidents effectively only with the density of crowd. Because the density of crowd is also very high in some normal circumstances, such as going school, working and so on. Therefore, this paper defines a feature called Blocking Rate of Crowd, which is used to detect the unexpected crowd incidents. It analyses some parameters which may affect the blocking rate of crowd. And then it gives the blocking rate of crowd in both normal scenes and abnormal scenes, which confirms the validity of this argument.At last it introduces common classifier in pattern recognition, which includes Back-Propagation Network. This paper decided to use Back-Propagation Network as classifier to detect unexpected crowd incidents. After discussion of how to build a Back-Propagation Network, it uses some samples to train the Back-Propagation Network. And then it uses the Back-Propagation Network to classify the remaining samples. This method is proved to be effective through these experiments.