Conclusion and Future Work

• With regards to strobing techniques, our research concludes that "intelligent" strobing (specifically selecting sprites to provide an even distribution with little overlap) does have potential.  One could imagine, for example, an optimization algorithm that selected sprites so as to minimze overlap while maximizing the number of chosen sprites.   At the same time, we have noted that intelligent strobing works in some situations better than others, and so an automated system should be interactive, letting the user choose a point on the continuum of timing vs. spatial clarity.  At the same time, more research needs to be done in encoding temporal information (e.g., speed) through alternative methods (like our first attempt at mapping velocity to opacity) in order to combat the flaws of straightforward intelligent strobing.
• Automatically generating comic strips is a daunting task.  We have only considered simple sequences with identical frames, reminiscent of Muybridge's photographs. In this case, the main research problem lies in determining which frames of the video represent "key poses," which seems tractable enough; shape similarity metrics could be used to compare silhouettes of the actor.  From there, the range of possibilities is large.  Adjusting the layout of frames on the page, changing the size and placement of the "gutters," and experimenting with non-rectangular panels are all part of the everyday life of the comic artist, yet the automation of these tasks are formidable research obstacles.
• We've discussed how some video sequences are more suited to strobing techniques and others to comics, and an interesting research project would be the formalization of these criteria and ways to automate it's detection.  By considering the amount of overlap in a given clip, a computer might be able to decide whether to create a strobing visualization or a comics-based one.
• Perhaps the most interesting research topics lie in the automation of motion indicators.  Our tool for generating motion lines was useful, but was still essentially a drawing program; it had no knowledge of the underlying images.  One could instead imagine a tool that loaded the video clip itself, upon which you could place markers that it would track backwards in time using optical flow. These paths could be used to automatically create artistic motion lines, or to define the direction of a motion blur streak.  As the two techniques - motion blur (from photography) and motion lines (from hand drawn art) - converge, we inevitably will see the creation of new motion indicators that were impossible to create in traditional media.