Abnormal Crowd Behavior Detection Based On Kinetic Energy And Rough Set

Nowadays, with the enhancement of population density in china, the probability of abnormal incident is increasing in the crowded public areas. In order to guarantee the peoples’ safety and minimize the economic loss, we must give an alarm timely and effectively when an abnormal event has happen. As an effective measure of social security, intelligent video surveillance system has become more popular. It extracts useful information from video automatically and processes the information. It can not only save a lot of manual work and financial resources, but also improves working efficiency. As a branch of intelligent video surveillance system, abnormal crowd behavior detection based on video has become a hot topic in the world.After studying the traditional algorithm of abnormal crowd behavior detection, we propose two kinds of new algorithm. One is based on improved kinetic energy, and another is based on rough set. The main research content includes the following aspects:We propose an algorithm based on improved kinetic energy for disperse events detection. In the traditional algorithm based on kinetic energy, the only factor to be considered is kinetic energy in the detection of the abnormal events. But in real life, there is always noise in video, which can lead to the kinetic energy of a frame to change suddenly. So system may make a misjudgment. Because of the above reason, we make the improvement to the kinetic energy formula. We not only consider the change of velocity, but also the change of crowd distribution. When disperse event occurs, the state of people usually changes from aggregation to decentralize, and the velocity will also become lager. So we combine crowd index of distribution with velocity for the feature in improved kinetic energy. Finally we can detect anomaly in the video by setting an appropriate threshold, we test our algorithm in UMN dataset and the video segments made by ourselves. We have done some contrast experiments to compare the proposed algorithm with the other algorithms. The experimental results show that the recognition rate of our algorithm is higher than other algorithms in the contrast experiment.In the existing algorithms, only a specific kind of abnormal event is tested and the different abnormal events could not be classified. So there are great limitations for these algorithms. We proposed an algorithm of abnormal crowd behavior detection based on rough set for testing scatter, fighting, sprint, which are more common in our life. Using our approach we can not only determine whether there is abnormal behavior in the video or not, but also can distinguish what kind it belongs to. On the other hand, as a mathematical tool of describing uncertainty, rough set do not need any prior knowledge about the objects to be classified. The advantages of our approach are that all of the decision regulations can be obtained by analyzing the video data using rough set and the subjectivity and randomness of observer can be reduced in the video data processing. Experiment on the videos made by ourselves shows the result that our approach is effective for detection of anomalous event.