Combining Multimodal And Temporal Contextual Information For Semantic Video Analysis