Slashdot Mirror
Researchers Develop Genuine 3D Camera
cylonlover writes "Cameras that can shoot 3D images are nothing new, but they don't really capture three dimensional moments at all — they actually record images in stereoscopic format, using two 2D images to create the illusion of depth. These photos and videos certainly offer a departure from their conventional two dimensional counterparts, but if you shift your view point, the picture remains the same. Researchers from Ecole Polytechnique Federale de Lausanne (EPFL) hope to change all that with the development of a strange-looking camera that snaps 360 degrees of simultaneous images and then reconstructs the images in 3D."
...Street View does?
Does anyone know if the Microsoft Kinect is classified as a "true" 3d camera under these criteria?
So I RTFA and WTFV , and the asshole at the computer put on some fucking glasses, I call shenanigans..
of Stereoscopic....it's polyscopic? I dunno...this still seems like more of the same.
Unless it's doing a lot of moving around, it's just stereoscopic on steroids. If it stays in a fixed position, even though it has more than two cameras, all the objects are at fixed points. Until it can accurately judge the height, width and depth of an object without faking it in reconstruction, or making an educated guess - it's just more of the same. Humans suffer from the same limitations, but they fix this by moving the viewpoint around until a coherent 3D image is constructed.
Unless you have cameras that can accurately measure objects and move in the X and Y dimensions enough to cover the entire scene and all viewpoints for a given object, you're stuck in the same position - educated guessing.
I never knew I was using such worthless vision capabilities until now.
I do hope to upgrade to the far superior bug eyes which will allow me truly see in three dimensions.
"You should always go to other people's funerals; otherwise, they won't come to yours." -- Yogi Berra
This contraption looks like a Dalek! With multiple eyes instead of the single eye-stalk, mind you, but Dalek nevertheless!
It may be 360 degree, but not 3d. It doesn't process depth any more than a traditional 2d camera, it just takes a wider angle of view.
The summary and article make it seem like this is some revolutionary *3D* device. It isn't. What it does to create 3D imagery has been around for a long long time (done in software, perhaps on a dedicated chip). The only newsworthy thing about this is that it can do very large panoramas.
...you still have your work cut out for you, blade runner.
Insert self-referential sig here.
You're right, it's nothing new. It's not even real 3D, it's "stereoscopy" to a higher degree. We were doing true 3D analysis 15 years ago; video analysis of gait and other motions on a treadmill. We used multiple cameras mounted orthoganally, and a digital mixer to combine and record onto SVHS tape w/a SMPTE time code. Post recording analysis was done using Peak Performance (brand) software. This was way before the cinematographers and game makers started doing this. There's little new in science.
Chaos maximizes locally around me.
Whoever tagged the story "france" got it wrong. The *real* Ecole polytechnique is of course in France, but this one is in Switzerland.
There's only one kind of "genuine" 3D camera, and it requires very special film and one of absolute stillness or high-power pulse lasers. We call the process "holography," and if it doesn't do that it's not a real 3D "camera."
Words mean things.
It sounds like this is a combination of the Kinect and the Google Street View or yellowBird camera. It had to read the article and watch the video twice, because initially it sounded like they were promising more than this could do. Turning a town or campus into a 3D model for a game sounds quite doable; you just need to move the camera around a ton as it records. As for getting a different perspective at a concert, he said you need several cameras. If you have a lot, then yeah, I can see a smooth perspective transition in real time being possible, but you would need a lot of these in the area.
Here's a 5 cam version already commercialized http://bit.ly/aHDafK
All the 3D cameras that use an array of 2D cameras, like this one does, have a huge fault.
It is a big inefficiency and inaccuracy in aquiring the image. A large quantum mechanical loss of information of 99% to 99.9%.
The problem can be simply stated as that the camera is 4D, because it is a 2D array of 2D cameras,
while the image is 3D. This means that the camera for a 3D picture of about 1000 pixels on a side must be 1000 times bigger than necessary, or have 1000 times longer exposure, or a combination of these.
SOLUTIONS: All solutions rely on making the camera 3D instead of 4D.
1. Take a short movie while changing focus on a 2D camera.
Thus time in the movie is the 3rd dimension, translated into different focuses.
2. Make a camera with multiple 2D sensors placed at different focal lengths, and somehow transparentified, perhaps with half silvered mirrors.
This more complex camera can shorten the exposure time significantly, if there is enough light.
A combinations of these 2 methods is often most cost efficient in money and exposure time.
Thise were old thoughts of mine, and I do indeed consider this to be prior art, and quite obvious as well.
I also consider it to be evidence that I am better at this stuff than almost all 3D camera engineers.
Kim0+, M.Sc. Measurement & Quantum Physics
Wohooo!! \o/
TFA gets it wrong, too... Sure, it may be great for immersive experiences, but it doesn't even address the question of 3D. For that, we are still stuck with holograms.
Enjoy life! This is not a dress rehearsal.
I'm glad to see people calling "bullshit" on this. I'm big into developments of PhotoSynth/Bundler/PMVS and other interesting 3D photogrammetry, so this is close to my heart. I'd just like to clear up a confusion though: "You can do a better job of 3D scene reconstruction with three or four Kinects in different places[...]" Unfortunately you can't, as the Kinects use structured light reconstruction, and the IR light patterns from multiple Kinects would confuse each other.
From the article & video, all I can see is a higher-resolution version of an Omnidirectional camera, which is very common in mobile robots. Such as this list of about 50 different types! "http://www.cis.upenn.edu/~kostas/omni.html"
They keep referring to the notion of depth being used, but unless there is some big technology that they completely forgot to mention in the article & video, it just does the equivalent of pointing a camera into a bowl shaped mirror, allowing you to see in all 360 degrees at once. eg: "http://en.wikipedia.org/wiki/Omnidirectional_camera"
That is quite different to say it is truly 3D, since it is still a 2D image without depth, just that its wrapped around a circle shape instead of rectangle shape.
The amazing thing is... My realtor must have been a genius because when we sold our house 4 years ago he had that very same camera take a picture of our living room...
After all, we're talking 3-D and not 2-D!
Bullshit. This is not genuine 3D. This is just stereovision using a lot of cameras demonstrated by a guy with a 'orrible french accent that talks a lot about what could be done but in fact they do not even come close to what this other guy built in a fraction of the time using a MS Kinect: http://www.youtube.com/watch?v=5-w7UXCAUJE
That video also makes it very clear why the fantasies of the french guy will never come true. At least not with that camera.
0x or or snor perron?!
This isn't what I want in a 3D camera. I want to be able to spin the scene after moving from original point. I want to see what's behind something. The cameras need to be encapsulating the scene.
this is bad advertisement. And timothy ought not have posted it.
As someone who has worked in stereoscopic research, there is nothing new to it in this 'development'. Except, of course, maybe the brute force real-time stitching of the images. The idea to arrange a multitude of cameras on a half-sphere has grown a beard over decades.
Worse, there is not much of a difference between a traditional '3D'-view (which isn't, actually, 3D), and this arrangement. A quarter century ago some chaps had a somewhat functional setup with 6 cameras and 5 perspectives. In these days, we can - thanks to advances in computers - calculate any intermediate 2D perspective with parallax. 'And what', is the only comment I could give.
And most relevant, probably, it doesn't address the most pressing question: How to project it; how to reconstruct a (calculated) 3-dimensional object in view-space (dunno if this word exists, but is the best construction I could think of)? And don't come and tell me, to use a similar setup of projectors! because that wouldn't work. Much ado about nothing. The good prof is eventually just hoping for tenure by advertising this thingy.
I do agree, the stitching algorithm could be new, more powerful, more precise. But then, the public wouldn't actually be impressed sufficiently.
No, EPFL, you didn't do much of a service to yourself with this clip, alas.
In engineering we use laser scanners that use a laser as a rangefinder to find out how far from the camera each pixel is. You then shoot from different perspectives to build a 3D scene that you can move around in. http://www.faro.com/3dimager/videos/
I love Jesus, except for his foreign policy.
"... they actually record images in stereoscopic format, using two 2D images to create the illusion of depth" My eyes work the same way.... dammit... if I want to see around something I have to actually move my damn head!
This is inverse 3D... instead of what we want to see from all the possible points of view, we have what is around the camera.in a plain view
I prefer the approach of the cameras surrounding the action in the first Trinity scene in Matrix.
Someone is still thinking in 2D.
I'm curious - isn't this an apples to oranges comparison? The camera is looking in all directions (i.e. panorama view) - but it isn't looking at the object from all directions (i.e. I still can't see something hiding behind the object). How does it "reconstructs the images in 3D"?
It seems to me that 3 flying cameras surrounding the scene should be able to capture practically any scene in 3D (4D, really, including time - though time might be a fractional dimension since it seems to move only forward).
--
make install -not war
3D Porn! Need I say more? Anyway now that I got your attention I remember sketching a camera like this 15 years ago and learning *then* that it was nothing new... I heard that Disney's panoramic movie at Disneyland LA used something like this, just in analog and without the processing. True 3d processing is like what you see at http://2d3.com/ ...
Thinkingman.com New Media
So let me get this straight...they invented a holograph??? wow.. I was making these (holograms mostly) back in high school. Does this give me prior art to sue them, make a million dollars, and buy drugs and whores?
This news is 2 weeks old. I saw this very same video on youtube while I was searching for 3d cameras some two weeks ago. Was just curious as to what format the 3d cameras, if there are any, were storing their images in...
I have been waiting for 3d cameras to arrive for a long time now. Was imagining something that will probably shoot some kind of lasers may be, into the scene and capture the depth of the objects that I point to and reconstruct the scene in 3d. I don't think the camera in this video is portable enough. Meaning, it won't go far. It will most likely be like fusion in the tea cup.
This was done years ago with two 360 degree panorama cams. The two spherical panoramic images gave reasonably convincing 3D imaging for a field of view of at least 160 degrees. This means you could dart your eyes anywhere in this field of view with acceptable results. To render subjects outside of this field of view, you had to reformat the subject at a particular area to match the two images well enough to allow viewing. This allowed viewing out to a FOV of nearly 180 degrees. Two pics. A bit of simple software to transform the spherical image to a panorama. Worked with video and stills.
Nice hobby project here, but the old way worked better in many ways. Lots of references on the web.
Because all the camera focal points are approximately at the same location, the images can be stitched together in software to create a full hemispherical view. This is essentially the same type of snapshot that is used in Google street view, and allows you to look in different directions.
In order to change position, however, requires depth information. Because the focal points are not EXACTLY at the same location, it is theoretically possible to estimate depth, although the practical reality is that the precision of these depth estimates are sensitive to image resolution and the ratio between baseline and depth.
Even if each individual camera was something like 14 megapixel, the accuracy of depth estimations would be so terrible as to be completely useless for anything more than a few feet away from the device. The author's neglect to mention this critical short-falling in their demonstration...
Point Grey Research has been doing this for years and years with their Ladybug spherical cameras (eg. http://www.ptgrey.com/products/ladybug3/Ladybug3_360_video_camera.asp). They have a much more innovative many-camera unit, too - it is a PCI-E device (weird, but high throughput!) with a square array: http://www.ptgrey.com/products/profusion25/index.asp