What’s the best way to control a computer using 3D gestures? One answer is that the best gestures are the most natural ones. But that leaves a puzzle: how do you determine the most natural gestures?
Today, Piotr Gawron and pals at the Polish Academy of Sciences in Gliwic, give us an answer of sorts. They’ve created a database of 3D human hand gestures and then “solved” the database to find a kind of ideal gesture.
They began with a subject wearing a motion-capture glove and then measured the way the glove moves while making 22 common gestures such as the A-OK sign, a walking motion with two fingers, shoving away an imaginary object and so on.
They then created matrix of the resultant data and found its eigenvectors, the mathematical solutions of a square matrix. They then turned these eigenvectors back into gestures to see what they look like.
These “eigengestures” are important solutions in the mathematical parameter space. But whether they are useful in the real world is not so clear.
The first eigengesture that these guys found is shown above. The sequence of images shows how the gesture looks from the point of view of the performer.
So what to make of this work? This new eigengesture certainly looks natural but Gawron and co say that most of the other eigengestures they’ve found aren’t natural at all. That may be related to their “negative finger bend parameters”!
But why should eigengestures be useful? There doesn’t seem to be any intuitive reason why this kind of analysis should produce useful or even natural gestures. It’s a little like measuring the shape of the letters in the alphabet, crunching the resultant data set in a similar way and suggesting that we’d be better off communicating with the “eigenletters” that it produces.
So despite an interesting approach, it looks as if eigengestures may have only limited use in developing the next generation of Minority Report-style interfaces.
Ref: arxiv.org/abs/1105.1293: Eigengestures For Natural Human Computer Interface