FUSION OF HUMAN EXPERTISE AND AI FOR PRIVACY PRESERVING CROWD MONITORING

Abstract

Accurate and prompt oversight of human behaviours is an essential factor for threat management within intelligent surveillance technologies. This article proposes a Privacy-Preserving Spatio-Temporal Graph and Transformer Network that handles the intertwined behavioural patterns in crowd data. For the surveillance environment, the dynamic network is assumed whereas light-weight cryptographic and attention-based concepts are introduced for enhancement of feature capturing and privacy shielding. Absence from the scene is predicted by combining the knowledge gained about behavioural habits and movement changes from activity data history. Based on experimental tests, the novel method is proved to be an effective way of handling complicated crowd situations when compared to other machine learning and deep learning techniques.