Here at the UIST conference I saw two talks today that both illustrated the same principle, in very different ways.
One was a talk about gaze tracking (sort of). It turns out to be very difficult for a computer to tell where a person is looking (which is frustrating, because this happens to be something our human brains do very well). There have been many approaches to the problem, most involving expensive and/or invasive hardware.
This paper started by redefining the problem to a simpler one: Is a person looking into the camera or not? It turns out that this much simpler question is quite solvable, with no special hardware required. The solution they showed even works with old photos or movies. And it happens to be good enough for a huge percentage of the applications for which you might think “gaze tracking” is needed.
The other paper showed a way to track someone’s hand position, using three simple and inexpensive photosensors instead of a fancy and expensive video camera. It can’t track as well as a camera. For example, you don’t get individual fingers and other finer details. But it turns out that those niceties are often not needed.
So here we have two ingenious papers, both illustrating the same principle: If you don’t know how to solve a hard problem, then solve an easier (but really useful) one.