Ontology-based hierarchical conceptual model for semantic representation of events in dynamic scenes

There is an increasing interest in semantic analysis of events in dynamic scenes in recent years, and many different methods have been reported for this challenging problem. A new approach towards event modeling and analysis with semantic representations is proposed in this paper. Our method is inspired by the entity-relation model in software engineering. It integrates all related information into a hierarchical conceptual model by the name of ontology, and defines events as significant changes and mappings of conceptual units in the mode. All concepts are represented by three basic components, an entity, a word, and a set of attributes. The lower level of our framework achieves the task of feature extraction, and in the upper level, semantically meaningful representations of events are received by using these words. So our framework is data-driven and provides semantic outputs. Semantic similarity measurement of concepts is another important problem. In this paper we propose a method that uses conceptual status vector (CSV) and weighted semantic distance (WSD) to deal with it. Experimental results are presented which demonstrate the effectiveness of our approach on real-world videos captured from different scenes.