Crowd synthesis based on hybrid simulation rules for complex behaviour analysis

The acquirement of video data for crowd anomaly detection and behavior analysis is a challenging practical issue due to the short of and deficiencies within real-life video footages. The construction of simulated crowd scenes using real actors is costly and often carrying potential safety hazards. In order to address this issue, a simulation or synthesis-based crowd video generation techniques is proposed and explored in this research through the investigation and integration of various behavioral models from physics, social science and human psychology fields. The investigation has been focusing on the separation and convergence behaviors among agents from different groups within a large crowd. This research also introduced an innovative crowd grouping model through adopting the concept of Velocity Perception based Social Force Model (VPSFM) and the Boids behavioral model. In the experiments the devised model successfully empowered a game-engine-driven crowd scene simulator that is capable of configuring and generating random crowd scenes of desired aesthesis, visual and behavioral realism. Furthermore, based on the proposed model, a defined grouping attraction force is proven effective when utilized to segment the randomly distributed and mixed crowds.