Crowd Flow Segmentation Using Lagrangian Particle Dynamics

Saad Ali, Mubarak Shah

Abstract: A crowd of people is composed of groupings that arise due to interdependence among its members. Advanced visual surveillance and monitoring capabilities for crowded scenes can make use of this inherent group-based composition of human crowds to understand its global motion dynamics and to compartmentalize it into sub-parts for detailed analysis. In this chapter we propose an algorithm that uses motion information to locate such distinct crowd groupings in terms of flow segments in videos of large dense crowds. The flow segments are located using a particle-based representation of the motion in the video. This representation enables detection of boundaries between dynamically distinct crowd groupings.