Slashdot Mirror
Fly Eyes for Spying Cameras
Roland Piquepaille writes "Even with sophisticated cameras, we can sometimes get poor pictures. This usually happens because cameras use an average light setting to control brightness. When parts of a scene are much brighter than others, the result is that you don't catch accurately all the parts. According to National Geographic News, by mimicking how flies see, Australian researchers can now produce digital videos in which you can see every detail. This technique could be used to develop better video cameras, military target-detection systems and surveillance equipment. Read more for additional pictures and references about these future surveillance cameras."
I find The first FA to be poorly written. It jumps between focusing [pun not intended] on two completely different concepts: dynamic range, and motion detection. The second article is slightly better.
We'll address dynamic range, since I know more about this aspect. The first page of the (first) article says he used "off-the-shelf components such as resistors, capacitors, and light sensors to build an electronic model". And then a sentence or so later says, "This would allow the camera to capture more complete images--such as, for instance, both the face of a person standing in front of a sunlit window and the scene outside." If you don't know much about digital imaging, let me just say that this is roughly the equivalent of "I used wheels and spark plugs to build a car and I now hope to win the Indy 500." The article is SORELY lacking in any real information about how he intends to extend dynamic range by using technology gleaned from flies.
There are several very real and working principles by which dynamic range can be extended, both unique to chip architecture (such as dual slope sampling) and implementable on a variety of chips (such as dual electronic shuttering). These are the types of things that it would have been cool for the article to discuss (imo). The second article at least includes a quote from him stating that fly eyes can adjust exposure independently.. this is a beneficial thing, and several CMOS imagers already exist that do this as well (i.e. dual slope operation, etc). You can also individually shutter pixels, or expose multiple frames per $interval (each with a different electronic shutter length) and then composite them.. however this last technique creates smear, which can be less than ideal, depending on your needs. I also know of a couple of patents for bayer masks that adjust individual pixel exposure in realtime (similar to those sunglasses that get darker or lighter) in order to compress dynamic range before it hits the CMOS/CCD.
One of the issues the articles really didn't get into at all, is storage of data. Higher dynamic range images require more storage space (as their bitspaces increase), and right now the major limitation in digital cinema and other similar realms is not imaging so much as writing all of the data to disk.. storage media speed (or cost/speed ratio, if you like) needs to do some catching up.
I am Jack's complete lack of surprise.
The problem, in short, is that digital sensors have pretty terrible contrast limitations. Film does too, to some extent, but with many years of experience these problems have been dealt with. You can only capture menaingful data within certain contrast zones. A good sensor may have 4 usable zones of contrast while your consumer digicam can probably only handle 2 and a half or three stops worth of contrast.
So what do you do? Well, since it's digital, take more pictures! expose the frame for a certain set of contrast zones and then repeatedly take the same shot with different contrast settings. Digitally combine the pics in Photoshop to render a frame with full contrast from the blackest black to the whitest white. The pictures look a little weird because we usually aren't able to see that much contrast rendered in Nature due to limitations of our eyes, but the results are pretty astounding.
After capturing the image, you need to display it somehow, or else there's not much point to the exersize.
Current screens and prints have a tiny dynamic range, on the order of 1000:1
So, once you've captured that image, where the brigthest pixel is a million times brigther than the darkest pixel, how are you going to show it ?
There's only one answer: compress the range, that is, map your numbers (in range 1 - 1000000) to much smaller numbers.
Problem is, now you've got terrible contrast in the midtones. The problem is that compressing the range compresses this part of the range too. So, assuming the monitor can display 1000 different brigthnesses, you end up with a picture where the brigthest pixel in a face is say 404 and the darkest pixel is say 397. Which makes the face essentially monotone.
A good place to start looking into this field is the Wikipedia entries http://en.wikipedia.org/wiki/Tone_Mapping and http://en.wikipedia.org/wiki/High_dynamic_range_im age.