Learning ontology rules for semantic video annotation