By using this site, you agree to our Privacy Policy and our Terms of Use. Close
JaggedSac on 27 April 2010
Edit Thread
WereKitten said:
JaggedSac said:

It looked like the bolded to me.

And the XY accuracy of Natal is around 3mm at 2m away from camera, based on the PrimeSense specs.  This would of course change if the FOV of Natal is different from the PrimeSense specs.  3mm is quite enough to make a real-time pointer, but it would not be pixel perfect.  I probably couldn't hold my hand still enough for pixel perfect anyway, as even the slightest shake would be displayed :)

It's really so short that I'm not sure about it being pointer or gesture based.

Anyway 3mm at 2m is the resolution of the camera. But the real resolution you could use for pointing with your hand would be the resolution of node positioning after the software processes the camera images to the 3d model. That's obviously quite a bit rougher, and we actually have some old numbers. I can't link to the original article, but didn't the main Natal dev talk to New Scientist about positioning limbs within a 4cm cube? That's probably the ballpark of your node resolution per frame.

Let's be optimistic and say that he meant a cube of 4 cm^3 volume and not a cube of 4cm per side. That would mean about 1.6cm linear resolution, ie if you map the screen to the position of your hand in a meter wide space, you get (1280/100)*1.6 pixel = about 20 pixel of error.

You can average n samples, of course, increasing the resolution by a sqr(n) factor, but adding lag for about n-1 frames. Thus you won't probably see anything better than 15pixels if we assume these are the numbers.

Two things.

What distance from the camera is he giving this estimation. ?

Secondly, the depth accuracy is around 2cm at the back end of Natal's range, this gives credence to that estimation.  I would hope that MS devs do not fuzzify the xy accuracy by a factor of 5 on the optimistic view, and 13 on the unoptimistic at 2m.