By using this site, you agree to our Privacy Policy and our Terms of Use. Close

Forums - Microsoft - Aptina to be supplier of 1.3-megapixel CMOS image sensors for new Xbox 360

Xoj on 23 July 2009
JaggedSac said:
Xoj said:
kowenicki said:
@xoj

they arent filming sports for hd broadcast... they are capturing movement and basic features of a face...... surely high end isnt required?

well if the sensors  are good maybe... sony partner with onmivision to make a almost free of noise ps3 eye also sony builts it own CMOS and CCD sensors.

noise signals can damage the input, and noise it likely to happen CMOS sensor, that's why it's cheap, for pictures you can postprocess and reduce noise but for videos it's hard, if the input have noise, output like going to be bad too, thats what happen in demo at E3 under not so great lighting condition caused a wrong output and the bone deformation you probably saw.

natal will require good lighting to work correctly, as it's now 1.3mpx CMOS (VGA 30fps or so does my HP pavilion 1.3mpx sensor.) it's not much better quality as Ps3 Eye that can even do in 120fps QVGA and 60fps in VGA. and the PS3 eye is 2mpx CMOS.

oh and for motion capture, studios built very expensive motion capture places.. that have tons of cameras to detect movement, and use high quality CCD sensors, and even that way sometimes it's a bit quirky and require  a person to fix a few things.

You do know that Natal shoots out its own infra red bursts.  Lighting conditions do not matter, at least for the motion control.  They even have video where they turn off the lights and play with Natal.  I am quite sure they do not need more than 1.3 million pixels of depth information to accurately capture multiple people's skeletons.

 

http://www.gamertagradio.com/vbportal/forums/showthread.php?p=69293

EDIT:  I also disagree completely that it was output from the depth detection device that caused the avatar issue.  It was almost definitely a software issue.  Since Natal can only detect depth of a two dimensional plane, the act of turning around needs some fancy algorithms to compute, not impossible, but it obviously needs work.  That is where the issue lies.

for facial recognization it's important, 1.3mpx is the sensor capacity only, recording size it probably less.

same as 2mpx eye can only do VGA 60fps.

while i agree input the camera was not the only issue was part of the problem, it's the CMOS technology which it's know to produce noise even with good light conditions a signal with noise can produce a bad output, especially with things like artificial intelligence and neuronal nets.

1.3mpx and CMOS are the cheapest mpx it's not the problem but CMOS sensor.