From visual information to knowledge : semantic video object segmentation, tracking and description