Learning rules for semantic video event annotation