Microsoft discussed the technical aspects of the the Natal motion capture system at the Consumer Electronics Show.
Alex Kipman, lead developer of the Natal, spoke with New Scientist about some of the technical hurdles they faced in designing the motion capture system. The Natal blasts infrared light out and based on how that light is reflected, makes a decision on what body parts it is seeing and in what position they are. The difficult part was teaching it how to interpret that data. The team collected terabytes of data on body parts that the AI uses to determine exactly what it is seeing. Because of all this programming, the Natal will consume 10-15% of the 360's processing power. The Natal will be released in November 2010.
"When we train this 'brain' we are telling it: this is the head, this is the shoulder. And we're doing that over millions of frames," says Kipman. "When it sees a new image it can tell you the probability it's seeing a certain body part based on that historical information."
Because of its huge knowledge of human anatomy, the Natal will be able to determine things it can't even detect. "It correctly positions your hand even if it's held behind your back," Kipman says. "It knows the hand can only be in one place."
The Natal will not ship with its own dedicated processor in order to keep the cost of the peripheral down. Therefore, Kipman and Co. (new sitcom!) had to figure out how to accomplish all of this computing without stealing too much processing power from the game developers. But is 15% of the CPU too much? Will Natal games look worse because of the peripheral leeching power? Only time will tell.
Source: New Scientist