Integrating Computer Vision Algorithms and Ontologies for Spectator Crowd Behavior Analysis

Capturing and understanding crowd dynamics is an important problem under diverse perspectives. From sociology to safety management, modeling and pre- dicting the crowd presence and its dynamics, possibly preventing dangerous activities, is absolutely crucial. In the literature, crowd has been classied un- der dierent categories depending on size and focus of attention. This chapter focuses on spectator crowd, namely that formed by people whose behavior is constrained by a structured environment, whose focus of attention is mainly shared, directed to a specic event. We rst propose the backbone of an on- tology of spectator crowd behavior based on a foundational analysis of both related literature and S-Hock, a massive annotated video dataset on crowd be- havior during hockey events. Then, we present a new methodological approach integrating ontological reasoning, performed with a new description logics based temporal formalism, with computer vision algorithms, allowing for automatic recognition of events happening in the playground, based on the behavior of the crowd in the stands.