DESCRIBING objects is so much easier when you use your hands, the classic being "the fish was this big".
For humans, it's easy to understand what is meant, but computers struggle, and existing gesture-based interfaces only use set movements that translate into particular instructions. Now a system called Data Miming can recognise objects from gestures without the user having to memorise a "vocabulary" of specific movements.
"Starting from the observation that humans can effortlessly understand which objects are being described when hand motions are used, we asked why computers can't do the same thing," says Christian Holz of the Hasso Plattner Institute in Potsdam, Germany who developed the system with Andy Wilson at Microsoft Research in Redmond, Washington.
Holz observed how volunteers described objects like tables or chairs using gestures, by tracing important components repeatedly with their hands and maintaining relative proportions throughout their mime.
Data Miming uses a Microsoft Kinect motion-capture camera to create a 3D representation of a user's hand movements. Voxels, or pixels in three dimensions, are activated when users pass their hands through the space represented by each voxel. And when a user encircles their fingers to indicate a table leg, say, the system can also identify that all of the enclosed space should be included in the representation. It then compares user-generated representations with a database of objects in voxel form and selects the closest match.
In tests the system correctly recognised three-quarters of descriptions, and the intended item was in the top three matches from its database 98 per cent of the time. Holz presented his findings at the CHI 2011 meeting in Vancouver, Canada, in May.
The system could be incorporated into online shopping so users could gesture to describe the type of product they want and have the system make a suggestion. Or, says Holz: "Imagine you want a funky breakfast-bar stool. Instead of wandering around and searching Ikea for half an hour, you walk up to an in-store kiosk and describe the stool using gestures, which takes seconds. The computer responds immediately, saying you probably want the Funkomatic Breakfast Stool-o-rama, and it lives in row 7a."