Gigapixel Camera Catches the Small Details

An anonymous reader sends this quote from Nature: "David Brady, an engineer at Duke University in Durham, North Carolina, and his colleagues are developing the AWARE-2 camera with funding from the United States Defence Advanced Research Projects Agency (abstract). The camera's earliest use will probably be in automated military surveillance systems, but its creators hope eventually to make the technology available to researchers, media companies and consumers. ... AWARE-2 sidesteps the size issue by using 98 microcameras, each with a 14-megapixel sensor, grouped around a shared spherical lens. Together, they take in a field of view 120 degrees wide and 50 degrees tall. With all the packaging, data-processing electronics and cooling systems, the entire camera is about 0.75 by 0.75 by 0.5 metres in volume. The current version of the camera can take images of about one gigapixel; by adding more microcameras, the researchers expect eventually to reach about 50 gigapixels. Each microcamera runs autofocus and exposure algorithms independently, so that every part of the image — near or far, bright or dark — is visible in the final result. Image processing is used to stitch together the 98 sub-images into a single large one at the rate of three frames per minute."

Comments

celebrate!

[Sparticus789]

I've always wanted an in-flight UAV to be able to diagnose me with skin cancer.

Meter-cube camera bricks

[mpeskett]

Coming soon to Minecraft?

Great...

[jmerlin]

Just when I thought I had plenty of space for my HD porn.

Only 3 frames per minute?

[Anonymous Coward]

Can't they just stack a shitload of gpus and parallelize the stitching? Where's the bottleneck?

Re:Only 3 frames per minute?

[timeOday]

RTA, they're working on a 10 hz version now. Although I think "stitching" is a bit misleading, since it normally implies compute-intensive pattern matching to register the images over each other. In this case it seems like you'd only do that once, since the lenses are fixed, and then just re-use the same mapping.

Anyways, that works out to about 18 MB per sensor per frame, times 100 (sensors) times 10 (hz) so 18 GB / s. So I think they'll need hardware video compression as well to do much with it. Otherwise you're filling a 4TB hard drive in a little over 3 minutes.

Re:Only 3 frames per minute?

[timeOday]

PS, obviously a hard drive cannot write that fast. Assuming 60 MB/s minimum sustained write speed, you'd need to stripe across 300 drives to push 18 GB/s :)

Re:Only 3 frames per minute?

[Alastor187]

Can't they just stack a shitload of gpus and parallelize the stitching? Where's the bottleneck?

Well considering 3 frames is about 3 GB of data, I would guess it takes that long to write the data to a harddisk. Accordingly, to the article they are still working on filtering out the data they do/don't want.

Re:Only 3 frames per minute?

[KingMotley]

Just to point out your errors: Your own link shows the deskstar having a 85MB/s minimum sustained write speed, and if you were to design a system that a minimum speed was required, you could quite easilyuse 2 drives, with one writing from inner track to outer, and the other from outer track to inner in a RAID-0, and as your graph shows, it would sustain 400MB/s between the two. You could also leave the inner tracks unused. Or use a faster drive that spins at 15k RPM. You still aren't near the speed you would need to write uncompressed images, but at least the numbers are realistic.

The obvious choice, use a SSD. My personal SSD (OCZ Revo 3 x2) can sustain 1.4GB write speeds without breaking a sweat. Save the images using standard jpg compression, and it should be extremely easy to do.

Re:Excellent

[PPH]

There! That's the one that did it! I'd recognize that pixel anywhere!

Big Brother upgrade

[J-1000]

I know this is inevitable and a nifty advancement, but it creeps me out. It's bad enough to have low-resolution cameras that can point in your general direction and maybe zoom in if they want to be extra nosy; now they'll have ultra high-resolution cameras with a 120 degree field of view that can identify you two blocks away.

Re:Big Brother upgrade

[girlintraining]

I know this is inevitable and a nifty advancement, but it creeps me out. It's bad enough to have low-resolution cameras that can point in your general direction and maybe zoom in if they want to be extra nosy; now they'll have ultra high-resolution cameras with a 120 degree field of view that can identify you two blocks away.

They already have cameras that can identify which brand of cigarette you're smoking from 80 miles away. Look up sometime. The problem isn't that the government has this technology, the problem is what they do with it.

Re:At last, my dream of a panoramic...

[girlintraining]

.....Roseanne Barr beaver shot at an exquisite level of detail is technically possible!

Just because a thing is possible doesn't mean it should be done. Also, eeeeeeeeeeewww.....

Woot! At last . . .

[dogsbreath]

a camera that truly matches the detail zooming capability of webcams on NCIS, CSI whatever and so on.

Why even stitch?

[2.7182]

What is the point of having the images stitched? Google maps doesn't need to. If the observer is going to look at the whole image then they can't look at it on a megapixel screen. The whole thing sounds fishy to me.

Re:What if they group-up 98 AWARE-2 cameras?

[LifesABeach]

Is this Camera the reason for the hold up on the iPhone 5?

Re:I've got a gigapixel camera...

[toadlife]

The GPs post and your reply are reminiscent of every vinyl vs CD debate/flame war I've ever seen.

In ten years, when we all have gigapixel cameras in our phones, are we going to have film devotees going on about how uniform pixels cannot possibly convey an image in the same way that film grains can?