Swarm-based motion features for anomaly detection in crowds

摘要: In this work we propose a novel approach to the detection of anomalous events occurring in crowded scenes. Swarm theory is applied for creation motion feature first introduced work, Histograms Oriented Accelerations (HOSA), which are shown effectively capture scene's dynamics. The HOSA, together with well known Gradients (HOGs) describing appearance, combined provide final descriptor based on both and characterize scene. Appearance features only extracted within spatiotemporal volumes moving pixels (regions interest) ensure robustness local noise allow anomalies small region frame. Experiments comparisons State Art (SoA) variety benchmark datasets demonstrate effectiveness proposed method, its flexibility applicability different crowd environments, superiority over currently existing approaches.