Abstract: In the mass of video resources, how to describe and represent video event content is one of hot issues in the current multimedia information processing. The original theory of case grammar for natural language understanding was extended; Case Frame was designed to describe the relationships between the structures of complex events in the sub-events. In Ref_Asso, which is one of relationships, spatio-temporal correlation was analyzed and reasoned among sub-events. In the experimental part, Case Semantic Frame Net was used to describe and understand the typical complex events in surveillance video. And results of users' retrieval were compared; in which video test set was described respectively with Case Semantic Frame Net and traditional Case grammar. Experiment results show that the new method can more accurately describe and understand complex events, and has a higher precision and recall rate in video retrieval.

Key words: case grammar, case framework, complex events, sub-event, Ref_Asso