Our work applies computer vision and image processing algorithms alongside an understanding of film language to develop tools that can be used in logging, archiving and indexing of video.
Members: John Mateer, John Robinson, Matt Day, Zhan Peng, Bop Dony (University of Guelph)
We are building reliable video parsing systems that analyse all types of motion picture footage and provide low- and mid-level descriptions. Based on earlier work in real-time projective transform estimation, we describe camera movement and robustly detect shot transitions. These techniques are incorporated in SALSA (see separate entry). They also form the basis for Reverse Storyboarding -- an automatic summarization method developed with Bob Dony of the University of Guelph, inspired by the methods of storyboard artists in the film industry. By applying work in face detection and tracking we have added descriptions of framings and groupings and we are continuing to research more sophisticated motion analysis in order to describe complicated camera movement and scene structure.
Members: John Mateer, John Robinson
SALSA is a system that analyses movies in a way useful for post-production and archiving. It detects shot boundaries, parses camera moves, and provides an efficient way for users to insert and edit higher-level annotations. It is the practical result of our wider research in motion-picture analysis described under the heading Motion-picture parsing and summarization. SALSA has no problem handling black-and-white footage, cartoons, or any other kind of source material. It is currently undergoing evaluation with film editors and other motion-picture industry professionals.