Silent_Buddha
Legend
Strikes me as an odd solution. We already have clear rules on what human movements are possible because of anatomical limits. If they have a fulled rigged skeleton, surely it's just a matter of mapping the joints to the locations found in the IR image? And because humans move at a relatively limited speed, you can expect any extreme changes from the previous image to be wrong. Yet from the sounds of it, and looking at the image, the points being tracked aren't actually attached to a skeleton system, and the skeleton is built up between them, or something. Rather than starting with a skeleton and mapping its joints to the point locations.
It isn't so much about tracking an absolute 100% relation of "skeletal points" to actual joints/locations on the human body. Natal is currently tracking what it "thinks" are most likely joints and whatnot, then out of a probability matrix chooses the points it "thinks" are the correct one that most likely correlate to a possible joint, etc.
The millions of images are to allow it to recognize more possible configurations, especially when you introduce clothing. IE to allow it to estimate with better accuracy where a possible joint may be buried under an unknown layer of clothing. Tight clothing probably doesn't pose much of a problem. But imagine extra baggy pants and maybe a baggy sweatshirt. A woman in a dress?
They combine that with established kinematics to then predict where those joints and skeletal points will be and how they will move if those points are obstructed at any moment.
Edit : Also the Kotaku article mentions a "50 MB software" solution. Is that reference data that needs to reside in RAM? If so, that's a 10% resource consumption tying in with the disclosed system requirements figure 10-15%.
Hmmm, interesting point. So if one were to assume that a 50 mb TSR is required on the console, it may be that the Natal unit is still doing the lions share of the processing.
Or in other words, we're still in blind speculation mode.
Regards,
SB