Although the technical task of extracting rigid, jointed components from monocular image sequences has been studied extensively, the process of automatically (or semi-automatically) exaggerating the motion is much more novel. This paper examines how to analyze real video sequences and to synthesize from these cues augmented versions of the sequences that exaggerate the motion represented. Performing this operation requires the solutions to two problems: estimating the two-dimensional (2D) motion of (here) rigid jointed objects, and, given this motion, introducing unrealistic cues to the motion sequence so as to enhance the motion presented to the viewer.
The approach presented here is to provide a collection of techniques, including ghosting, streak-lines, and motion caricaturing, that can be used by an animator. A critical human element in the process of introducing these exaggerations is to limit the exaggerations to those portions of the image sequence that should be exaggerated, and to restrict the exaggeration from portions of the motion sequence that should remain grounded.
Although the solution to the segmentation of objects into segmented rigid components is essential to the problem at hand, I found the portion of the paper that discusses synthesizing novel sequences to be much more interesting. I would have preferred the paper to have discussed this second problem in more depth, and to have pointed at the large amount of literature addressing motion estimation for rigid jointed objects for the first problem.