Real time temporal segmentation of MPEG video

One of the main customers of video annotation applications are TV broadcasters, and one of their key requirements is on-line annotation of received digital channels in order to offer dynamic value-added content-based services. For this purpose, apart from high levels of success in video temporal segmentation - a highly resource-consuming task previous to video annotation - a faster than real-time performance is also crucial. The paper describes a video temporal segmentation module able to detect abrupt transitions and all types of gradual transitions in real time, with levels of recall and precision not previously reached for so large and varied a set of video sequences.