Microsoft has a hit on its hands with Kinect, a motion-sensing camera for the Xbox 360, but it’s not keeping the secrets of the technology to itself.
A new video, courtesy of New Scientist (and embedded above), shows how Kinect’s camera first grabs a rough outline of the person standing in front of it, and then evaluates every pixel to separate the image into body parts.
From here, Kinect guesses the position of the user’s joints, allowing the system to generate top-down and side views of the subject as he or she moves around in three dimensions. This is an important distinction between Kinect and Sony’s Playstation Eye camera, which only detects images in 2D, and relies on the Playstation Move motion controller to track 3D movement. The Xbox 360’s body-recognition algorithm runs 200 times per second — roughly 10 times faster than previous techniques.
The New Scientist reports that Microsoft researchers are now trying to figure out how to identify joints directly, instead of detecting body parts and then estimating joint location. This would, of course, allow for even more accurate motion control — you know, just to make Dance Central a little less forgiving.