Tuco
I got Tuco'd!
- 45,420
- 73,489
Virtual Reality
- Thread starter spronk
- Start date
Skanda
I'm Amod too!
- 6,662
- 4,506
Brahma
Obi-Bro Kenobi-X
- 11,945
- 42,254
I don't think this is in 3D, but still pretty cool. The future in all this looks pretty awesome.
https://www.kickstarter.com/projects...-hd-360-camera
http://www.360.tv/gallery- Even does video.
Szlia
Member
- 6,560
- 1,318
Filming at 360? is one thing (a pretty trivial one too), but how do you record the difference between the two eyes for every position the user can take? That seems straight up impossible, which means you need to fake it with software (probably generating a 3D model of the scene - by picture analysis or with another sensor - and map the video on it), but that would require a whole lot of interpolation / time cloning (showing at a time t stuff that was recorded before or after t) which is guaranteed to result in a clusterfuck if the scene has any kind of structural complexity or movement.
Examples: you film a travelling by a fence. If the user turns his head toward the fence, it's game over because you are forced to time clone and have partial and/or inconsistent volumetric data. Now have a guy walk by and the complexity of the geometry of a face along with the time cloning needed means he will most likely look like a Picasso, especially if he is walking and talking. I hope they buy nice cars with their millions, but that stuff is not going to happen except in extraordinarily controlled environments with a volume of data captured outside filming (like scanning the face of the guy that walks by, etc).
On top of that, the end result is amusement park material and not much more.
Examples: you film a travelling by a fence. If the user turns his head toward the fence, it's game over because you are forced to time clone and have partial and/or inconsistent volumetric data. Now have a guy walk by and the complexity of the geometry of a face along with the time cloning needed means he will most likely look like a Picasso, especially if he is walking and talking. I hope they buy nice cars with their millions, but that stuff is not going to happen except in extraordinarily controlled environments with a volume of data captured outside filming (like scanning the face of the guy that walks by, etc).
On top of that, the end result is amusement park material and not much more.
You bring up a good point, that camera that they showed in the article doesn't measure depth, so in effect it's just a 360 camera like Google's Street View. Sure, maybe you can pan around a game/event in real time, but you probably wouldn't get that sense of depth that you would in a 3D game (haven't used Oculus yet, just basing this off what I've heard/read).
Tuco
I got Tuco'd!
- 45,420
- 73,489
Binocular 360 degree fov is a very challenging problem that hasn't been solved yet. Translational tracking from fixed cameras will never happen to any degree that people will want it.
For sports events like football the effects of 3d can be faked with more ease than say, boxing, because due to a small IPD our ability to perceive depth between something at 30 meters and 50 meters is not great. However with boxing where the particpants are between 1m and 7m away, you really have to be accurate.
A lot of games will probably switch to a rendering system where after a certain range (30-60 meters), the scene is only rendered once for both eyes, and only entities in the close range are rendered twice.
I'm pretty big on VR, but I'm a little dubious of the value of VR sporting events where you're placed at centerline of the pitch or behind the batter or whatever and can look around. Even with perfect VR (which we don't have by any means) you're not going to capture the essence of being at the game, but you are going to lose the ability to be automatically switched between the huge number of cameras they're using (62 at the last superbowl!). You could say they could use 3d fish eye cameras and you'd use the Rift to view what they see, but it's really disorienting to be in a moving camera.
I think the best you could hope for with VR sports watching is something like the cinema simulators they have, where you can watch games on a massive screen. The massive screen being displayed prevents you from being disorientated as it moves around, but you still have a larger screen than you have at home.
For sports events like football the effects of 3d can be faked with more ease than say, boxing, because due to a small IPD our ability to perceive depth between something at 30 meters and 50 meters is not great. However with boxing where the particpants are between 1m and 7m away, you really have to be accurate.
A lot of games will probably switch to a rendering system where after a certain range (30-60 meters), the scene is only rendered once for both eyes, and only entities in the close range are rendered twice.
I'm pretty big on VR, but I'm a little dubious of the value of VR sporting events where you're placed at centerline of the pitch or behind the batter or whatever and can look around. Even with perfect VR (which we don't have by any means) you're not going to capture the essence of being at the game, but you are going to lose the ability to be automatically switched between the huge number of cameras they're using (62 at the last superbowl!). You could say they could use 3d fish eye cameras and you'd use the Rift to view what they see, but it's really disorienting to be in a moving camera.
I think the best you could hope for with VR sports watching is something like the cinema simulators they have, where you can watch games on a massive screen. The massive screen being displayed prevents you from being disorientated as it moves around, but you still have a larger screen than you have at home.
Thinking about it a little more technically, there's a parallax between each camera in that sphere. However, since the distance between each camera is small, the depth accuracy isn't going to be very good. Stereo imaging on a static platform isn't that great for 3D localization. Humans are really good at it because we move our heads around to change perspectives which improves your depth perception. You could probably get a crude representation of something close to the camera, but it's not going to be anywhere near the fidelity of say the demo with a bunch of 3D people standing around:
plz don't embed youtube videos of naked people, especially in the sample image -tuco
Multi-Sky 3D Scan Tools - Oculus Rift - YouTube
By increasing the number of cameras and the spacing between them, you could get better 3D stereo localization... but to do that in real time is probably going to take a heck of a lot more processing power.
Well... I guess for a sporting event, you could delay the 360 camera feed to give you time to correlate points between each camera. So maybe it isn't that technically difficult?
plz don't embed youtube videos of naked people, especially in the sample image -tuco
Multi-Sky 3D Scan Tools - Oculus Rift - YouTube
By increasing the number of cameras and the spacing between them, you could get better 3D stereo localization... but to do that in real time is probably going to take a heck of a lot more processing power.
Well... I guess for a sporting event, you could delay the 360 camera feed to give you time to correlate points between each camera. So maybe it isn't that technically difficult?
Sorry about that softcore YouTube embed.
That's a great example, but you can see how involved the processing is to get that smooth hyperlapse -- and yet it's still not perfect. Watching the video, you can see lots of structure pop-in, especially during the rock climbing hyperlapse. I mean, that's a really hard problem, the rocks all look very similar, and so you're trying to match similar rocks together. Still, the result is super impressive and almost makes me want to get an Oculus to play around with it from a video processing perspective.
edit: The first half of their problem would've been made a lot easier if they had 6D info to begin with (xyz, roll, pitch, yaw) as opposed to estimating it. Doing it without that data makes their technique a lot more involved.
That's a great example, but you can see how involved the processing is to get that smooth hyperlapse -- and yet it's still not perfect. Watching the video, you can see lots of structure pop-in, especially during the rock climbing hyperlapse. I mean, that's a really hard problem, the rocks all look very similar, and so you're trying to match similar rocks together. Still, the result is super impressive and almost makes me want to get an Oculus to play around with it from a video processing perspective.
edit: The first half of their problem would've been made a lot easier if they had 6D info to begin with (xyz, roll, pitch, yaw) as opposed to estimating it. Doing it without that data makes their technique a lot more involved.
jeffvader
it's only castles burning
- 402
- 33
Siliconemelons
Avatar of War Slayer
- 10,790
- 15,130
Where I work we have an innovation lab- I would love to get a rift for there... if anyone is board of their DK2 or upgrading and needing to offload their DK1 we can do a donation receipt and you can get your tax credit etc. if you are into that kinda thing and could use some tax credit let me know!
Lanx
<Prior Amod>
- 60,662
- 132,792
anyone mind if i talk about the google cardboard, if you don't know what it is, it's this
tldr(see): get fake occulus rift with cardboard a few lenses and smartphone
you guys think this is a detrement or good thing for VR?
i decided to try it, not cuz i'm cheap, but for 20bucks why not, i'm gonna upgrade my s3 anyway, so i needed to find a use for it.
the raw materials (lens/magnets) come out to 15-20bucks, you can order a kit from amazon (which is what i did cuz i have prime free shipping) for 25bucks
Amazon.com: I AM CARDBOARD? 45mm Focal Length Virtual Reality Google Cardboard with Printed Instructions and Easy to Follow Numbered Tabs (WITH NFC) (Box Color): Cell Phones Accessories
so you essentially pay 5bucks for pre-cut cardboard. you can also just buy the raw materials kit, then goto the website, print out the pdf and trace it out arts and crafts style, eh.
interesting thing is that google has probably been doing this for years, they're now prepping google street view to work with an app called views, basically taking all of the street view data, the app is already working with google earth.
tldr(see): get fake occulus rift with cardboard a few lenses and smartphone
you guys think this is a detrement or good thing for VR?
i decided to try it, not cuz i'm cheap, but for 20bucks why not, i'm gonna upgrade my s3 anyway, so i needed to find a use for it.
the raw materials (lens/magnets) come out to 15-20bucks, you can order a kit from amazon (which is what i did cuz i have prime free shipping) for 25bucks
Amazon.com: I AM CARDBOARD? 45mm Focal Length Virtual Reality Google Cardboard with Printed Instructions and Easy to Follow Numbered Tabs (WITH NFC) (Box Color): Cell Phones Accessories
so you essentially pay 5bucks for pre-cut cardboard. you can also just buy the raw materials kit, then goto the website, print out the pdf and trace it out arts and crafts style, eh.
interesting thing is that google has probably been doing this for years, they're now prepping google street view to work with an app called views, basically taking all of the street view data, the app is already working with google earth.
Tuco
I got Tuco'd!
- 45,420
- 73,489
I'd say it's pretty neutral toward VR. From what I hear the experience granted by the cardboard VR is just enough to get a taste of VR but doesn't cross the tremendous threshold of people wanting to spend a lot of time with it. Even the DK2 doesn't cross that threshold except with a few games.