Was that the projected pattern solution, and the TOF camera was just shored up to prevent its use by other parties?
Old Discussion Thread for all 3 motion controllers
- Thread starter betan
- Start date
- Status
LightHeaven
Regular
Lag does seems pretty minimal with one person in front on the camera, but when the second player showed up it got pretty jerky...
If that's the performance they are getting for 2 players that pretty much scraps off 4 simultaneous players at once.
Probably for that, and for ensuring technology for future use, when the costs go down.
If that's the performance they are getting for 2 players that pretty much scraps off 4 simultaneous players at once.
Probably for that, and for ensuring technology for future use, when the costs go down.
Sorry, guys, I find interesting the latest discussion on this but I would like to read your opinions on the questions in my previous post, too. If those questions were discussed before, please provide a link.
Thank you.
Thank you.
I'm no expert, but I don't see why not...other than the lighting conditions and potential overhead...in fact I'm sure before eyetoy first came out I saw similar demos at ECTS
Intuitively, deducting poses from 2D data should be very doable, applying some amount of assumptions and context. We can easily recognize the way people stand with one eye closed, or indeed looking at a 2D photo. Human vision is complex, but it shows that source data with depth removed is still sufficient.
For posterity, if you can recognize the face, you can infer the front of the body because people generally can't turn their heads more than 70~80°. If you assume symmetry of the torso, you can infer the angle (and some relative depth information). You can assume symmetry of arms, joint constraints and constant lengths to "train" your system to an individual. As the arms appear thinner or thicker, longer or shorter, you can infer changes in depth and angles etc.
One could argue that depth info makes the processing easier, as it resolves ambiguities. I don't know how much of a factor that really is though. Some ambiguities can already be solved with assumptions about the joint structure (a knee will never tilt forward and such). Background separation is maybe the big thing here, but it was never demonstrated if the tech actually does that well -- the demo stages were always wide and empty; demo setups were all sideways, far away from the next "back" wall.
I have seen some work in 2D skeleton tracking. I think you may be able to find them on youtube.
Here's a paper on a possible approach:
http://dircweb.king.ac.uk/papers/Martínez del Rincón09_32358224/bmvc_abstract.pdf
(from Google search result. Not sure if applicable to gaming)
EDIT:
On a related note, a new studio has been formed to do motion games for the consoles:
http://www.joystiq.com/2010/04/15/side-kick-founded-to-work-on-motion-based-games-for-next-gen-ga/
I got a really annoying popup earlier, so here's a Youtube link to the Natal onstage demonstration: http://www.youtube.com/watch?v=xwRtMcUrlJQ
The thing that irks me is that just like all the older CES demos from 3DV, etc, the joint markers are jumping around a bit. And I'm not just talking about when there are 2 people on screen at once.
The thing that irks me is that just like all the older CES demos from 3DV, etc, the joint markers are jumping around a bit. And I'm not just talking about when there are 2 people on screen at once.
Move and the next generation
I was considering the next generation of interfaces and how they relate to next generation consoles. With Nintendo they've obviously made their bed 100% with their Wiimote so they aren't confliced between two competing interfaces and with Natal Microsoft has a completely seperate identity for the camera and its seperate and/or used in conjunction with the standard console controller. However with Sony and the move its use requires at minimum between 3 and 5 accessories for a single person.
How can Sony reconcile the use of Move whilst maintaining their current controller layout for regular games which require it and without breaking the bank in terms of overall expense? Their current model is half in Wiimote world and half in the regular controller world and for the average person buying a console next generation thats a huge number of accessories. The layout for a mote style controller is nothing like the current Dual Shock 3 and they had a huge backlash from people when they showed off their prototype boomerang controller.
I was considering the next generation of interfaces and how they relate to next generation consoles. With Nintendo they've obviously made their bed 100% with their Wiimote so they aren't confliced between two competing interfaces and with Natal Microsoft has a completely seperate identity for the camera and its seperate and/or used in conjunction with the standard console controller. However with Sony and the move its use requires at minimum between 3 and 5 accessories for a single person.
How can Sony reconcile the use of Move whilst maintaining their current controller layout for regular games which require it and without breaking the bank in terms of overall expense? Their current model is half in Wiimote world and half in the regular controller world and for the average person buying a console next generation thats a huge number of accessories. The layout for a mote style controller is nothing like the current Dual Shock 3 and they had a huge backlash from people when they showed off their prototype boomerang controller.
GamesRadar posted their impressions of the publicly available Move games to date:
http://www.gamesradar.com/f/hands-on-with-playstation-move/a-2010041510586912018
http://www.gamesradar.com/f/hands-on-with-playstation-move/a-2010041510586912018
corduroygt
Banned
That motion fighter bit confirms my suspicions. Full body tracking is not viable since most gamers do not have the muscle coordination of athletes or superhuman game protagonists. I mean if I record myself jumping and climbing, I'm not going to look as elegant as Drake, by a long shot 
They'll all have to resort to some sort of gesture system.
They'll all have to resort to some sort of gesture system.That's just one game. It'll be on a case-by-case basis depending on the vision of the developer. Because the game has you doing complex pro moves, it's going to have to be gesture-based.That motion fighter bit confirms my suspicions. Full body tracking is not viable since most gamers do not have the muscle coordination of athletes or superhuman game protagonists. I mean if I record myself jumping and climbing, I'm not going to look as elegant as Drake, by a long shot
3DV made a more down-to-earth boxing game that tracks the 1:1 location of where your fists land. So it is possible to do a fighting game without gestures, but only if you want to simplify the game and those gameplay mechanics that make Motion Fighters a more complete fighting game.
But I too hope Sony isn't getting too concerned with cinematic feel, and rather starts tapping some of the controller's potential.
By bundling the PSEye and PSMove with the PS3. That way you could buy a DS3 and use it was the NAVCON (R) or as a standard controller.
Thank you for your reply.
I see your point.
I also thought that some kind of preliminar calibration should help the system to recognize and locate the joints (like, "stand this way like the silhouette").
Then, I come again to the same point: skeletal tracking relies in a clever software (mostly) and Microsoft knew how to be the protagonist, here, taking a role that Sony could have played before and showing off their product as something revolutionary.
NATAL's hardware only provides a depth map, and I dare to say that those images are still 2d material. If it featured 2 cameras, then we could speak about real 3d input. It mades me think that PSEye can achieve similar results, at the end of the day, with proper software. I also suppose that an efficient use of the SPU's could help a lot on this.
Did I say too much? Maybe someone with a more technical approach can correct me on this.
Thank you.
I read the second link, yesterday, but I can't read the paper (the web browser just loads a bunch of random letters and symbols...).
ShadowRunner
Veteran
Skeletal tracking on a 2d camera is possible but will only be skeletal tracking in 2d still. A 2d camera cant tell if a leg is moved forward, to the camera it hasnt moved at all.
Your simply not going to want to do it on a 2d camera alone much better of using face tracking(which can provide depth info from distance between two eyes) and the two wands(also depth info), as in the puppetry demo and then us 2d skeletal tracking to fill in the gaps.
Your simply not going to want to do it on a 2d camera alone much better of using face tracking(which can provide depth info from distance between two eyes) and the two wands(also depth info), as in the puppetry demo and then us 2d skeletal tracking to fill in the gaps.
I think one of the Move games use body tracking (not necessarily skeleton tracking). The GamesRadar article has some details.
Skeleton tracking is probably one of the many approaches to reason about your movement (e.g., What is the gamer doing right now ?). Other approaches may use constant motion (like EyeToy games), color matching, motion controller data to approximate the gamer's action. These approaches may be used together to achieve better results.
Skeleton tracking is probably one of the many approaches to reason about your movement (e.g., What is the gamer doing right now ?). Other approaches may use constant motion (like EyeToy games), color matching, motion controller data to approximate the gamer's action. These approaches may be used together to achieve better results.
No. It's not a 3D spacial/volumetric capture (which is 3D scanning a la medical and model-creation), but it is capturing 3 positional dimension values. Sterescopic depth perception isn't the only way to determine distance to an object, but it is the natural way so gains an unfair reputation.
As for depth camera capabilities, put your hand flat upon your chest. Now move it a few cms forward and twist it ever so slightly to one side. The size and shape of the hand on a 2D image will be very hard to track and determine that the hand has moved forwards, whereas a depth camera will have exactly that info. It elliminates completely the lighting issues of contrast-based image recognition, which were a considerable inconvenience to EyeToy and EOJ. Move solves this by providing a light source, which just goes to show relying on a camera will not provide robust results. There's a reason TV studios are full of bright lights!
Though the theory of image-based human skeleton tracking is sound, I consider it highly implausible a solution could be found in the consumer space. Such technology ought to first appear in movies, where real actor tracking without blue-screening and point markers for mo-cap would be a huge advance. If it's not being used there yet, I doubt it'll appear first in a games console.
I know. Thank you. I already have Adobe Acrobat and its plug-ins for IE and Firefox, but still... Weird.
Maybe here it could be useful the preliminar callibration which I talked about. Once the system has recognized the length of a thig, for instance, any further changes in that length should be interpreted as rotation. Other parameters, such as the position of a foot, can help to interpret if the rotation is forward or backwards.
Yes, I remember this demo and your explanation is correct. Thank you.ShadowRunner said:
Yes, I see the difference.
Another example of a difficult position/rotation to track (for a standard camera) is when a leg is rotated forward/backwards, isn't it? Is the same example that I said above in response to ShadowRunner. Do you consider that the possible solutions that I pointed are likely to be considered?
Then, you don't "believe in NATAL"? Sorry if I didn't understand.
- Status
Similar threads
