Researchers devise a system capable of analysing a stream of video and producing a text-based description of its content