Eulerian Video Magnification
June 4, 2012 2:17 AM   Subscribe

A new video processing technique can amplify subtle changes in color and faint movements to show previously barely perceptible changes - like a heartbeat.
posted by Zarkonnen (30 comments total) 42 users marked this as a favorite

 
The example where they can monitor a newborn's heartrate visually just by over-amplifying the color changes in the frame is completely nuts. I'm guessing this sort of processing will help with things like intelligent security camera monitoring and the like.
posted by disillusioned at 2:28 AM on June 4, 2012 [7 favorites]


Coming soon to a casino near you!

But how well will it work with low quality input?
posted by clorox at 2:56 AM on June 4, 2012


And how well will it work on Dick Cheney?
posted by Zarkonnen at 3:11 AM on June 4, 2012 [3 favorites]


ENHANCE
posted by datter at 3:27 AM on June 4, 2012 [15 favorites]


The part right at the end of the movement caused by the mirror moving in the camera is great. Non destructive stress monitoring of objects would be possible. These movements coupled to an FE model for model inputs would be interesting. Very precise displacement field inputs .
posted by stuartmm at 3:52 AM on June 4, 2012


The throbbing in the wrist part made me spontaneously say "cool!" out loud. A million and one practical applications for this, of course, but I can also see people using it to make some really interesting art.
posted by Mizu at 4:10 AM on June 4, 2012


Quite amazing. A whole world of information is out there, we just don't see it naturally.
posted by gen at 4:36 AM on June 4, 2012 [2 favorites]


The heartbeat one is shown as a medical usage but of course we already have monitors for that. It's a dog whistle to intelligence agencies that they can do "lie detector" type analysis using only video now.
posted by DU at 4:52 AM on June 4, 2012 [7 favorites]


A whole world of information is out there, we just don't see it naturally.

Yeah, my second thought was "this is how aliens are going to see us". "Why do they keep alternating between red and yellow?"
posted by DU at 4:52 AM on June 4, 2012 [2 favorites]


Personally, I keep alternating between red and yellow because it feels gooooooooooood.
posted by Flunkie at 5:03 AM on June 4, 2012


Fascinating!
posted by rmmcclay at 5:44 AM on June 4, 2012


How long until this becomes the new polygraph?

On preview: What DU said.
posted by The 10th Regiment of Foot at 5:45 AM on June 4, 2012


And people are always HISSING. And GURGLING. And they STINK. Gross. So Gross.
posted by seanmpuckett at 5:45 AM on June 4, 2012 [1 favorite]


The example where they can monitor a newborn's heartrate visually just by over-amplifying the color changes in the frame is completely nuts.

It's even cooler when you realize a large number of cardiovascular related problems might show themselves months if not years sooner if you're able to identify the fact that a "pulse" might be different on different sides or parts of the body.
posted by Blue_Villain at 6:03 AM on June 4, 2012 [10 favorites]


Holy mackerel that's neat stuff.

I also assume it will show up in porn first.
posted by jquinby at 6:58 AM on June 4, 2012


That is amazing stuff. I don't really have much else to add. It makes me wish I'd gone into math or physics, despite my utter lack of interest and innate talent.
posted by Scattercat at 7:14 AM on June 4, 2012


You are all thinking about this backwards. The money isn't in adding motion, it's in removing it: Now you can have a steadycam done entirely in software. Your typical point'n'shoot will no longer give you as many blurred images. The cheapo zoom lens on your new DLSR (with this software implemented) will be nearly as good as the super-high-end image stabilized one.

Also lie detectors, but that will be brushed aside in the small print because STEADYCAM. YouTube shots of the tornado will now look like they were produced on a Hollywood set...
posted by caution live frogs at 7:29 AM on June 4, 2012


There is an app for this - Philips makes an app to monitor heartrate and breathing using nothing more than iPad / iPhone camera and sensors.
posted by borborygmi at 8:00 AM on June 4, 2012


Yay! Surveillance!
posted by regicide is good for you at 8:18 AM on June 4, 2012


Holden: So you look down you see a tortoise. It's crawling towards you.
Leon: Tortoise, what's that?
Holden: Know what a turtle is?
Leon: Of course.
Holden: Same thing.
posted by hanoixan at 9:01 AM on June 4, 2012 [1 favorite]


I wanted to see a motion-amplified video of somebody dancing, or at least walking.
posted by Pruitt-Igoe at 9:53 AM on June 4, 2012 [1 favorite]


The money isn't in adding motion, it's in removing it

No, it's about using something as relatively inexpensive as a camera as a more intensive sensor device. Want to detect subtle movements in a scene? Attach bright markers to the object in question, film with a steady camera, and process in software in order to amplify. No need to get 50 small sensors, just 50 small pieces of plastic and a single camera. Etc.
posted by suedehead at 11:31 AM on June 4, 2012


The explanation was simple enough that I was thinking "yeah, yeah, big deal", but the results were amazing. It's the most surprising thing I've seen in quite a while.

The biggest down side is that writers of cheap police procedurals will feel free to abuse "Enhance" even more.
posted by benito.strauss at 11:39 AM on June 4, 2012 [2 favorites]


Hmm... One downside is that you have to know what frequency to look for. For example, with the heart rate, they found it by tuning their parameters to look for frequencies that corresponded to what you would expect for a heart rate.

The other interesting thing, though will be doing this with high-speed video, looking for patterns that happen really quickly.
Your typical point'n'shoot will no longer give you as many blurred images.
Uh, no. This works by analyzing subtle changes between frames, rather then in-frame.
posted by delmoi at 12:24 PM on June 4, 2012


How hard would it be to look at the power spectrum and find the two or three biggest "bumps", and show the results of the original algorithm run with each of these settings? (There's no reason you have to produce just a single output images.)
posted by benito.strauss at 12:46 PM on June 4, 2012


I think some people are missing a key factor (else I am): This works by sampling over longer periods of time. It wouldn't necessarily be able to see the pulse of everyone walking past a shop window, but it would be possible to see the pulse of a sleeping baby (as shown) or a person sitting still for an interview, because the periodic nature of the changes is happening so many times that it can be filtered, averaged, and amplified. If you only had a few samples (i.e., window shopper) you wouldn't have enough data to filter and amplify.
posted by odinsdream at 2:07 PM on June 4, 2012 [3 favorites]


The enhancement of the ulnar pulsation in the wrist was cool. I could imagine a smartphone app using that technique to aid arterial blood sampling.
posted by ppl at 4:56 PM on June 4, 2012


fantastic stuff. Also potentially gives phenomenological insight into the perceptual consequences of certain drugs.
posted by spacediver at 8:37 PM on June 4, 2012


For those interested in getting fewer blurry images from their cheap cameras, this is the tech demo for you.
posted by tylerkaraszewski at 8:54 PM on June 4, 2012


I just hope they can go back to and add this into old films. I'd love to know which scenes from ESB Mark Hamill had a boner in (spoiler alert: all of them.)
posted by seansbrain at 11:52 PM on June 4, 2012


« Older Eduard Anatolyevich Khil, aka Mr. Trololo, has die...  |  The Motorcycle Collection Muse... Newer »


This thread has been archived and is closed to new comments