Crowd Behavior Analysis in Video Sequences

The analysis of human activities in crowded scenes is the most challengeable tasks in computer vision. Tracking and understanding individuals actions in dense scenes is a problem till yet not be fully solved due to occlusion between objects.
The new area of interest in computer vision is the crowd behavior analysis and modeling. Broadly speaking, there are two levels of crowd analysis: 1) individual level and 2) global level. At the individual level, the goal is to extract and understand behavior of each moving object in the crowd. At the global level, the goal is to model the behavior of the group as a whole. In both cases, one can perform behavior understanding and anomaly detection by analyzing motion features and characterizing so-called “normal behavior”. In contrast, detecting “anomaly” or “abnormal behavior” refers to the action of locating activities that do not conform to “normal behavior” or fall in its respective labeled class.
In this project we will focus on the modeling of crowd-flow solutions without tracking. After the modeling of crowd behavior, we will be able to detect high-level abnormalities such as traffic jams, crowd of people running amok, etc.