PlayStation Move technology thread

Sony has a 1 lens 3D technology that they use in their stereo cameras, they even have a full body motion controller like Kinect called ICU maybe they will use that for next gen.

 

Kinect isn't dependent on the amount of light reflected, but the distortions of a known pattern, which, although affected by different amounts of absorption, are a far more robust solution to a straight intensity evaluation. If sophisticated software could be developed to map and track areas of different intensity, eg. calibrating to a standard pose, with fast enough refresh it could work, but would likely be prone to contrast issues like EyeToy. This is where project patterns and pulsed light come in, which are probably covered completely by patents.

 

Yeah, Dr. Marks mentioned (in one of the interviews) that Sony has filed for this patent while working on PSEye. He discovered it accidentally when he saw his own video on a TV screen even though all the office lights were turned off. Turns out that the PSEye is sensitive enough to see him under low light condition (i.e., illuminated by light from the TV screen).

I presume you need to be quite close to the box, may be like a laptop ?

 

they can track the depth by size like they do with the Move and face tracking and the Playstation Eye already work in the dark only using the light from the TV.






Edit: I said kinect will have the same problem because some materials don't reflect IR light. and if it's black the system will just see it as part of the background.

 

They're tracking the face there, which provides several points of contrast always present against a predicatable background - eyes and mouth contrasting with skin tone. Full body tracking a la Kinect would require 'seeing' the torso and limbs. A pale top against a background throws up issues. EyeToy failed to detect my hand contrasted against a pale-blue wall in bright conditions as there was not enough contrast to register the pixels as different within the threshold needed to accomodate noise. If PSEye were to read my hand as forwards, if I then move it to one side and similarly coloured pixels are behind, how does it know that the pixels occupying that position are background and not hand moved back? Given human physiology, it would be possible to determine what's happened as clearly the hand can't move back 3 metres to where the wall is in one frame, but you'd need some funky code-voodoo to pull of a robust full-body tracking system achieved optically from a single camera. The best I know of so far is an 8 camera motion capture system. There's a YouTube vid of a 3 camera system on PS3 Linux, but no skeleton to see how effective it actually is.

 

Okay that makes sense. I posted some back of the envelope calculations in the old thread stating I didn't think it was possible to get sub-millimetre depth accuracy without u/s, based on the camera resolution and distance between camera and controller.

I was wondering how they were doing it as I didn't think it was possible, if it turns out they aren't then it all makes sense!

 

this is why I said if they used a IR emitter & a lense cap that only let the IR in so it would only be seeing the IR light that's reflecting back from your body.

and even without that they can use dynamic background extraction so it can know your hands from something that's in the background.

I never said it would be perfect I just asked could it be done.

edit:



see the part at about the 2 minute mark when he was reaching his hands into the scene I'm talking about if they used a IR emitter to do that on a bigger scale.

 

Background removal is tricky, as Sony themselves said. There's a good example in Kung Fu live (starts 2:30) that shows the blobbyness of background removal. Very similar to the point-cloud silhouettes of Kinect.

Without a projected pattern, recreating Kinect in PSEye with a filter cap as you suggest won't work IMO. Light intensity alone won't cut it, which is why every company doing this either has multiple cameras or some projected pattern.

 

http://next.videogame.it/playstation-move/video/6320/#media


PlayStation Move: Video Interview with Richard Marks ( next.videogame.it)

 

Thanks, great interview! It was a lot of work but the information is so dense in that interview and one of the best I've seen so far, so I wrote it out (hopefully noone else did that yet in English or I'll feel silly, it took me an hour ).

 

Sony PlayStation Move: what the future holds

http://www.techradar.com/news/gaming/sony-playstation-move-what-the-future-holds-702750?artc_pg=1

 

Wow, you should leave the entire content on techradar so they can make some dough from the traffic.

 

LOL ok

 

http://iwaggle.blogspot.com/2010/07/look-at-some-third-party-playstation.html

 

I can't see any of the above images and video, onQ. Am on an iPad. YouTube videos should work.

EDIT:
Forgot the China government blocked YouTube.

 

The RTS controls look really good. The idea of painting units instead of drag selection is really good. And the speed with which you could zoom in and out seems ideal for the genre.

 

There are other interviews kicking around with Dr Marks and Anton that are interesting, although not necessarily very technical. (the hiphopgamer interview is funny even for me, nowgamer (I think) have a more 'financial/strategic/less graphically interesting' interview).

Looking at the GT racing wheel, does anyone else think that maybe those recent racers could be patched for move support? (I think they were very high budget, and a 'partial-re-release' along with sony marketing at a better time of the year might reduce their losses?)

 

For the wheel attachment, wouldn't the eye lose track of the ball when the wheel is turned? If they put the trackball at the top instead of the side, at least you could turn a decent amount without it getting lost behind the base, but the way they have it now, wouldn't it lose track for left turns? Oh well... I love my DFGT.

 

The eye tracks the glowing ball, and finds the linear position of the Move controller. "you moved up 8 pixels, left 7.2 and back 1cm".

For a steering wheel, you want the rotational position of the controller - provided by the gyroscope/magnetometer inside the Move.

i.e. you could unplug the eye and technically the move would still function as a steering wheel to the same of degree accuracy.