GROUP BEHAVIOUR PROFILING FOR DETECTION OF ANOMALY IN CROWD

Groups are the main entities that form the crowd. So understanding group level properties is vital and practically useful in a wide range of applications, especially for crowd abnormality detection. This paper aims to address the problem of modeling different behaviors captured in surveillance videos for the applications of normal behavior and abnormal behavior detection. A novel framework is developed for automatic behavior profiling and anomaly detection based on the clustering based group analysis. These behaviors can be effectively applied to public scenes with variety of crowd densities and distributions and are potentially important in many applications like crowd dynamic monitoring, crowd video classification and abnormal event detection in security surveillance.