Fool me, this is real good shit ! [the VR thread]
- Sir_LANs-a-lot
- Sheepie Whore

- Posts: 12143
- Joined: Mon Mar 10, 2003 1:00 am 0
- Location: Here
Re: Fool me, this is real good shit ! [Oculus Rift]
I a little confused - What happens if you turn your head? Doesn't all the stuff move with it?
[Sir_LANs-a-lot]
Meat is murder.
Tasty, tasty murder.
Meat is murder.
Tasty, tasty murder.
- Sir_LANs-a-lot
- Sheepie Whore

- Posts: 12143
- Joined: Mon Mar 10, 2003 1:00 am 0
- Location: Here
Re: Fool me, this is real good shit ! [Oculus Rift]
I suppose this is a little more practical/consumer friendly than their "IllumiRoom" concept.
[Sir_LANs-a-lot]
Meat is murder.
Tasty, tasty murder.
Meat is murder.
Tasty, tasty murder.
- Deckard
- Sheepie Whore

- Posts: 10748
- Joined: Fri Jun 18, 2004 11:27 pm 0
- Location: Falmouth, Cornwall, UK
Re: Fool me, this is real good shit ! [Oculus Rift]
I am sure I read somewhere it stays where it is as if it was fixed in the roomSir_LANs-a-lot wrote:I a little confused - What happens if you turn your head? Doesn't all the stuff move with it?
Now with a new Mac Mini Pro so dont judge me !
- pogo
- Sheepie Whore

- Posts: 14180
- Joined: Tue Sep 10, 2002 8:28 pm 0
- Yahoo Messenger: BrassSparrow
- Location: County Durham
Re: Fool me, this is real good shit ! [Oculus Rift]
That’s another problem. How do you generate a 3D model of what you see good enough to virtually interact with. State of the art for this is probably mo-cap but that depends on two things. The virtual model is hand built. If it uses e.g. textures from the real world they are applied by hand to the model. And the capturing is done by decorating the thing being scanned with many dots and using many cameras.gravy wrote:So an initial scan could have higher latency but after that orientation ends up being a much simpler pattern matching exercise on a data set that has good temporal coherence
Outside of such an artificial and tightly controlled environment I don’t know what state of the art is but I suspect it’s not taking a picture of your coffee table on the fly and turning it into a workable 3D model. And your walls, and your floor. It’s one thing to demo this in a tightly controlled environment, probably custom made, which has been input to the device. But generating it on the fly from a typical front room?
Re: Fool me, this is real good shit ! [Oculus Rift]
I don't know what you're on about. You only need geometry information from the environment and the stuff you overlay is just standard computer graphics.pogo wrote:That’s another problem. How do you generate a 3D model of what you see good enough to virtually interact with. State of the art for this is probably mo-cap but that depends on two things. The virtual model is hand built. If it uses e.g. textures from the real world they are applied by hand to the model. And the capturing is done by decorating the thing being scanned with many dots and using many cameras.gravy wrote:So an initial scan could have higher latency but after that orientation ends up being a much simpler pattern matching exercise on a data set that has good temporal coherence
Why would you need mo-cap?
- pogo
- Sheepie Whore

- Posts: 14180
- Joined: Tue Sep 10, 2002 8:28 pm 0
- Yahoo Messenger: BrassSparrow
- Location: County Durham
Re: Fool me, this is real good shit ! [Oculus Rift]
Mo-cap is one way people today accurately take real world stuff and put it into computers. Apart from that I don’t know what state of the art is; I recall they’ve been working on getting computers to recognise objects from photos but it’s still pretty primitive. There are 3D scanners but they require rather more, such as dedicated integrated lights. Here’s one which claims to be the fastest 3D scanner
http://www.creaform3d.com/en/metrology- ... -goscan-3d
Typical (i.e. small) objects are scanned in 5 minutes or less, and that’s five minutes of moving the scanner around it from the video, bathing it in light.
http://www.creaform3d.com/en/metrology- ... -goscan-3d
Typical (i.e. small) objects are scanned in 5 minutes or less, and that’s five minutes of moving the scanner around it from the video, bathing it in light.
Re: Fool me, this is real good shit ! [Oculus Rift]
Mo cap is motion capture, it has nothing to do with scanning what objects look like or converting 2d still images to objects. You can get laser scanners for that but again it has no bearing on this which is about; overlaying CG onto the real world.pogo wrote:Mo-cap is one way people today accurately take real world stuff and put it into computers. Apart from that I don’t know what state of the art is; I recall they’ve been working on getting computers to recognise objects from photos but it’s still pretty primitive. There are 3D scanners but they require rather more, such as dedicated integrated lights. Here’s one which claims to be the fastest 3D scanner
http://www.creaform3d.com/en/metrology- ... -goscan-3d
Typical (i.e. small) objects are scanned in 5 minutes or less, and that’s five minutes of moving the scanner around it from the video, bathing it in light.
- pogo
- Sheepie Whore

- Posts: 14180
- Joined: Tue Sep 10, 2002 8:28 pm 0
- Yahoo Messenger: BrassSparrow
- Location: County Durham
Re: Fool me, this is real good shit ! [Oculus Rift]
It needs a fairly accurate 3D model of the world to do that as well as seen in the video/as well as described by the journalists.gravy wrote:Mo cap is motion capture, it has nothing to do with scanning what objects look like or converting 2d still images to objects. You can get laser scanners for that but again it has no bearing on this which is about; overlaying CG onto the real world.
Re: Fool me, this is real good shit ! [Oculus Rift]
You just need two images so you can calculate the edge-parallaxes. It's pretty much what Kinect does already.pogo wrote:That’s another problem. How do you generate a 3D model of what you see good enough to virtually interact with. State of the art for this is probably mo-cap but that depends on two things. The virtual model is hand built. If it uses e.g. textures from the real world they are applied by hand to the model. And the capturing is done by decorating the thing being scanned with many dots and using many cameras.gravy wrote:So an initial scan could have higher latency but after that orientation ends up being a much simpler pattern matching exercise on a data set that has good temporal coherence
Outside of such an artificial and tightly controlled environment I don’t know what state of the art is but I suspect it’s not taking a picture of your coffee table on the fly and turning it into a workable 3D model. And your walls, and your floor. It’s one thing to demo this in a tightly controlled environment, probably custom made, which has been input to the device. But generating it on the fly from a typical front room?
Re: Fool me, this is real good shit ! [Oculus Rift]
The Kinect projects a room full of IR dots so it has known reference points. Thats a little easier then doing it dynamical from no known points.
---
That said right now, on tablets, theres quite a few bits of software that will look at a table in realtime with no markers on it, and overlay 3d models on it aligned - some even to the extent they can go behind real objects.
Tracking is poor, registration slips a lot....but it is also running of relatively low power systems with nothing but a camera.
This stuff typically works by looking at high-contrast areas of the environment, defining recognisable points, then using trig to build a 3d mesh from those points as you move about. The more you move the better model of the world it can get as "noise" will slowly get eliminated.
My guess is Microsofts system will have a 360 camera, or cameras at a few angles around the device, and use something like this for absolute positioning. It will have gyros and accelerators as well to help with relative displacements.
---
That said right now, on tablets, theres quite a few bits of software that will look at a table in realtime with no markers on it, and overlay 3d models on it aligned - some even to the extent they can go behind real objects.
Tracking is poor, registration slips a lot....but it is also running of relatively low power systems with nothing but a camera.
This stuff typically works by looking at high-contrast areas of the environment, defining recognisable points, then using trig to build a 3d mesh from those points as you move about. The more you move the better model of the world it can get as "noise" will slowly get eliminated.
My guess is Microsofts system will have a 360 camera, or cameras at a few angles around the device, and use something like this for absolute positioning. It will have gyros and accelerators as well to help with relative displacements.
-
want to read eye-wateringly bad stories? No? Ok.
want to read eye-wateringly bad stories? No? Ok.
Re: Fool me, this is real good shit ! [Oculus Rift]
No it doesn't.pogo wrote:It needs a fairly accurate 3D model of the world to do that as well as seen in the video/as well as described by the journalists.gravy wrote:Mo cap is motion capture, it has nothing to do with scanning what objects look like or converting 2d still images to objects. You can get laser scanners for that but again it has no bearing on this which is about; overlaying CG onto the real world.
It needs to find rudimentarty geomtery to place shit on (table tops etc) and depth information for the scene you're looking at. That's basically a big bitmap with a z value for each pixel.
They're saying that it's the same tech kinect which can generate a depth buffer of what you're looking at really quickly.
Re: Fool me, this is real good shit ! [Oculus Rift]
It cant be that though, Kinect is a fixed device looking at moving things. It can get information by projecting IR dots and looking at how they change.They're saying that it's the same tech kinect which can generate a depth buffer of what you're looking at really quickly.
If a HMD projected dots both the dot projector and the camera move in the same way - you gain no information.
It could be a time of flight sensor, but more likely just optical I think.
-
want to read eye-wateringly bad stories? No? Ok.
want to read eye-wateringly bad stories? No? Ok.
Re: Fool me, this is real good shit ! [Oculus Rift]
I think it's a HMD kinect sensor. The X360 is structured light based but the XBOX ONE doesn't work like that, it's time of flight.Darkflame wrote:It cant be that though, Kinect is a fixed device looking at moving things. It can get information by projecting IR dots and looking at how they change.They're saying that it's the same tech kinect which can generate a depth buffer of what you're looking at really quickly.
If a HMD projected dots both the dot projector and the camera move in the same way - you gain no information.
It could be a time of flight sensor, but more likely just optical I think.
Re: Fool me, this is real good shit ! [Oculus Rift]
ah, wasn't aware of that. I thought time of flight were generally too pricey.
-
want to read eye-wateringly bad stories? No? Ok.
want to read eye-wateringly bad stories? No? Ok.

