"Synthetic Tracking" Makes It Possible to Find Millions of Near Earth Asteroids 101
KentuckyFC writes "Astronomers think that near-Earth Asteroids the size of apartment blocks number in the millions. And yet they spot new ones at the rate of only about 30 a year because these objects are so faint and fast moving. Now astronomers at the Jet Propulsion Laboratory have developed a technique called synthetic tracking for dramatically speeding up asteroid discovery. Insteads of long exposures in which near-Earth asteroids show up as faint streaks, the new technique involves taking lots of short exposures and adding them together in a special automated way. The trick is to shift each image so that the pixels that record the asteroid are superimposed on top of each other. The result is an image in which the asteroid is sharp point of light against a background of star streaks. They say synthetic tracking has the capability to spot 80 new near Earth asteroids each night using a standard 5 metre telescope. That'll be handy for spotting rocks heading our way before they get too close and for identifying targets for NASA's future asteroid missions."
What's the size of an apartment block? (Score:1)
How many hedgehogs to an apartment block?
Re: (Score:2)
Roughly ten thousand olympic swimming pools.
Re: (Score:2)
Yet noticeably smaller than Wales.
Re: (Score:2)
Blue Wales, or Pilot Wales?
Is it just me? (Score:5, Interesting)
Or does the submitter not see the apparent logical flaw in the way the described this process. If you're going to line up each image so that the asteroid is a single sharp pixel and the stars are streaks, doesn't that suggest that you already know which pixel is the asteroid? In which case you don't really need to search for that particular asteroid, no?
At a minimum the submitter or the editors need to think whether their description of the procedure is good.
Re: (Score:2)
Yes and no. I was thinking the same thing but, what if you do the process multiple times, shifting in different directions each time, and possibly again with different amounts of shift per frame? Then you have maybe 8 or even 80 different results to look at, which could be weeded out by algorythmically rejecting any that contain no bright spots.
Re: (Score:2)
An apartment block would be about .04 furlongs across. You should be able to figure it out from there.
Re: (Score:2)
Ah, but I was wondering about volume in hogsheads [traditionaloven.com].
Re: (Score:2)
All units below are based off an average newborn baby:
B = Baby (Avg.
Re: (Score:2)
That's exactly what they're doing: brute force analysis. All directions and all "reasonable" variations of speed for a given series of spot pictures, then filter for spot intensity. Decent use for super computing.
Re: (Score:1)
No, rather a complete waste of supercomputing, since there are already better algorithms for doing the same thing.
Re: (Score:2)
Care to share?
Re: (Score:1)
Yeah, because the scientists at JPL are all idiots compared to some random douchebag on Slashdot who thinks he knows everything.
Re: (Score:1)
Re:Is it just me? (Score:5, Insightful)
Article doesn't have a good description.
My guess is you take a bunch of timelapse frames of the same sky.
Then you overlay them at offsets in different directions which would keep any moving objects in the same place.
Picture doing 36000 sequences of overlays:
360 degree variation in 0.1 degree increments at 10 different radial velocities
Most of those sequences will just show blurred gray washout, but if you happened to hit the right direction as a moving object at the right speed, your overlaid image sequence will effectively keep the moving object in the same spot of the frame, which will result in the average brightness for that pixel or pixles to be higher than the surrounding blurs.
Just a guess...
Re: (Score:2)
But you lose a point for not converting distance to international standard apartment block scale.
Re: (Score:2)
Interesting idea, I wonder if that's ever been tried? I guess the feasibility depends on what the angular motions of these objects are.
In this case though, they "simply" take a lot of short-exposure images of the same region and add them together. From the abstract:
The technique relies on a combined use of a novel data processing approach and a new generation of high-speed cameras which allow taking short exposures of moving objects at high frame rates, effectively ``freezing'' their motion. Although the signal to noise ratio (SNR) of a single short exposure is insufficient to detect the dim object in one frame, by shifting successive frames relative to each other and then co-adding the shifted frames in post-processing, we synthetically create a long-exposure image as if the telescope were tracking the object with a significantly higher SNR.
Re: (Score:2)
Actually, rereading the abstract, it seems that GP was spot on:P
Re: (Score:2)
Interesting idea, I wonder if that's ever been tried?
See Inverse Synthetic Aperture Radar [wikipedia.org]. It's basically just taking it one step further. Where ISAR starts with the assumption that you know where something is, and thus know how to shift the captures to produce the synthetic image, this brute forces the computation and waits for a sufficiently strong signal to show up.
Comparison? (Score:2)
Re: (Score:2, Informative)
Article doesn't have a good description.
Before the detection of the NEA, its velocity vector is unknown. However, we find this vector by conducting a search in velocity space. To do this we have developed an algorithm that simultaneously processes the synthetic tracking data at different velocities. The velocities searched initially have (x,y) components that are multiples of 1 pix/frame in each direction. This is a computationally intensive task: for example, the shift and add process for 120 images for 1,000 different velocity vectors requires over 1011 arithmetic operations. However, with current off-the-shelf graphics processing units (GPU) with up to 2,500 processors and teraFLOPS peak speeds, we were able to analyze 30 sec of data in less than 10 sec. Once the NEA is detected in this initial search, an estimate of velocity becomes possible. Using this velocity we refine the astrometry relative to a reference star in the field and determine the velocity to a much higher precision. Elsewhere we plan to describe the details of the synthetic tracking algorithm and report its performance, including its false alarm rate.
Re: (Score:1)
For anyone who wants to look into this, the above is a description of a standard computer vision algorithm called the generalised Hough transform. Its one of those really old techniques that has stood the test of time and remains useful for real problems today.
Re:Is it just me? (Score:4, Interesting)
"If it were moving at speed v, it would show up when I shifted the pictures by x pixels." Repeat for likely ranges of v, watch for bright spots. No contradictions required.
Re:Is it just me? (Score:5, Informative)
From TFA:
The difficult part of this is knowing which way to shift each image. Shao and co have solved this by brute force: they take consecutive images and examine all possible shifts to see which resolves the fast moving asteroid.
That's a lot of computation (they try 1,000 different velocity vectors), but that's what computers are really, really good at.
Re:Is it just me? (Score:4, Interesting)
Sometimes the best way to solve a big problem is to just get a bigger hammer.
I had a problem once that I could probably have solved using some very pretty, complex, elegant formula. But after examining the problem space, I figured I could brute-force it, with a basic fitness algorithm, in about 2 seconds. The overall process it was a part of took between 90 and 300 seconds. The other benefits were, it was quite readable, and didn't require any advanced math or knowledge of the problem to see what was being done. The fact that the pretty formula would have improved performance at most about 2% made it an easy choice.
Re: (Score:2)
Congratulations. You have discovered how thinking beings solve problems. Consider this planet a computer brute forcing the answer not to emergence of sentience, but also to the most ironic way to become extinct.
Re: (Score:1)
As the material in the second link states, computationally they do a brute-force approach, trying all possible directions and speeds of drift of the object, then look at any resultant images that show a single object. A nightmare from a computational complexity approach, but with the amount of HPC horsepower out there these days, a workable one.
Re: (Score:1)
they probably just use a quad SLI setup with Nvidia GPU's to do this. Something they'd be perfectly adequate for
Re: (Score:2)
Re: (Score:2)
href=https://www.trustedsec.com/february-2011/building-the-ultimate-bad-arse-cuda-cracking-server/>Something akin to this.
although there certainly are updated versions available, for large operations. This was from 2010, but I couldn't find a more recent picture.
Re: (Score:2)
Re: (Score:3)
Yes the description may be flawed but seems to me that this is essentially a technique that has been in use for a long time, at least back to the 1950's, with systems like surveillance radar where several ping round trips are superimposed (added together). Involves delaying/storing the received signal and adding back together in a time correlated manner. Noise tends to reduce and object reflections tend to reinforce resulting in an effective improvement in the signal to noise ratio. In the early days,
Re: (Score:2)
STANDARD UNITS, PLEASE (Score:2, Funny)
Wha...? Asteroids the size of apartment blocks?
Can we please have this measurement in a standardized unit, like Volkswagen beetles?
Man. I thought Slashdot was going downhill back when it was mostly a CueCat fansite, but this really takes the cake.
Re:STANDARD UNITS, PLEASE (Score:4, Insightful)
Re: (Score:2)
Re: (Score:2)
For me, it was the ponies.
Oh wait.
Re: (Score:2)
Which is worse, using 'apartment blocks' or 'metre' for measuring.
Is a '5 metre telescope' really a standard?
I have a couple of Meade telescopes but not a 'Metre".
Re: (Score:3)
Can the mass be measured in '73 Beatles full of Cüe Cats?
Re:STANDARD UNITS, PLEASE (Score:4, Insightful)
Ahh but are you talking about VW Beetles from the 1950s and 60s or the fat ones from the 70s? Or perhaps you're talking about the "New Beetle?" or now the "New, New Beetle?" I just want to get some specifics here so I can make sure my bunker can withstand a 20 MegaBeetle impact.
really? (Score:1)
Re: (Score:2)
UPS truck drives into river: Tuxedo Guy holds sewing machine.
Re: (Score:1)
Yes. From the article:
"Before the detection of the NEA, its velocity vector is unknown. However, we nd this vector by conducting a search
in velocity space. To do this we have developed an algorithm that simultaneously processes the synthetic tracking
data at dierent velocities. The velocities searched initially have (x,y) components that are multiples of 1 pix/frame
in each direction. This is a computationally intensive task: for example, the shift and add process for 120 images for
1,000 dierent velocity vect
Re: (Score:2)
You missed the "synthetic" they take lots of static pictures with the same point over time.
Then use a computer to skew the images it in all the directions and speeds and do the search.
Re: (Score:2)
Sounds like this http://en.wikipedia.org/wiki/Synthetic_aperture_radar [wikipedia.org]
Why not use optical flow (Score:2)
It seems like it would be much simpler to just use optical flow to find moving objects.
Re: (Score:2)
Stacking images doesn't just magically produce data from nothing. It is entirely possible to threshold all of the bright objects and/or logarithmically scale the samples and look for motion in the low amplitude region close to the noise floor.
Thank you NSA (Score:1)
Quit spamming for "medium.com" (Score:5, Informative)
"Medium.com" is one of those aggregator sites. Don't link to them. Link to the actual paper [arxiv.org]. Thank you.
They had to use the Palomar 200 inch telescope to make this work. There aren't many big telescopes in the world, and they're booked months in advance. They got a few hours of observing for one night, and good results. But they'd need a lot more observing time on big scopes to do their survey.
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
And yet.. (Score:3)
It always amazes me that the people that complain all the time about Slashdot with dupes, bad articles, etc. come back every day just so they can tell everyone how bad it is. It's as though they sit and wait for it just so they can make long winded comments on how bad Slashdot is. It is getting really old.
Isn't there a simpler way? (Score:1)
Why would you brute-force the pixel shifts to get the asteroid in a single spot, when the fixed background stars are already a stable reference for co-aligning the images? It seems the simpler way (even with the short exposures) would be to cross-correlate and co-align to the background stars, then look for the "dotted" path of the moving asteroid. And isn't it already done this way?
Re: (Score:2)
I can only guess that in a single image, the asteroid is indistinguishable from noise, and even overlaying the images doesn't bring it out clearly enough to be spotted. Perhaps, once you do all the meta-math, it turns out to be easier just to do many overlays at multiple assumed velocities (which can be filtered down to what we expect of these asteroids) and then look for a bright spot.
Stacking images is simple. Spotting lines of dots in amongst the noise could be fairly tricky to do with any kind of conf
Re: (Score:2)
Now do that easily and in an automated fashion for an unknown moving object which is scarcely brighter than the noise level and often orders of magnitude dimmer than your reference points (background stars). The amazing part isn't finding a moving object, it's finding a near-invisible moving object in a rather automated fashion.
'standard 5 meter telescope'? (Score:2)
Re: (Score:2)
Re: (Score:2)
Well, hardware shops with really big shelves in the corner, anyway.
Let's all just agree (Score:1)
Re: (Score:3)
Re: (Score:1)
Great news, but may be pointless... (Score:1)
They weren't doing this already? (Score:1)
Re: (Score:2)
Up until this point it was too expensive computationally. Each set of images require 10^11 computations to process, a non-trivial amount of processing even for a big cluster of GPUs. Asteroids are really, really dim, and small ones are even harder to detect. The signals that they're trying to process are below the noise threshold.
It is called stacking, and already done (Score:3)
The method of overlaying multiple short images so that the asteroid is a pinpoint additive composite of multiple images and the stars become trails is not new.
The latter technique is called "stacking" (a word existing for quite a long time and meaning the same as their "synthetic tracking"). It is regularly done to image and get astrometry on faint objects, when speed and direction of movement are already known (e.g. in follow-up observations on a Near earth Asteroid that already has some observations over the previous hours/days and hence a preliminary orbit). That part is really not new, and there is no need to invent new terminology ("synthetic tracking") for it.
Frankly, it is weird that the authors nowhere mention "stacking" as an existing technique that is often used in imaging faint asteroids. It suggests they did not investigate whether their "new" technique is really that new. Yes, they innovate on it, but they did not invent a completely novel technique.
Re: (Score:3)
Re: (Score:1)
This approach appears to be slightly different from traditional frame stacking, in that they are utilizing a low read noise (really, 1e- doesn't pass the smell test--there's got to be some tradeoffs) to take a large number of frames with short exposure times. The only other interesting approach they are taking is searching the velocity space, for which (given 1000 points) they need a 2500-node HPC cluster (you do have one of those in your closet, yes?). From their description, they are also only searching 1
Re: (Score:2)
It's worth keeping in mind, though, that you don't publish CalTech papers and get time on the Palomar 200" by being a dim-witted slacker.
True. I think their presentation of things is more the result of current publishing demands: useful or even innovative is not good enough anymore to get your paper through, it needs to be "new" and "never done before" instead of an innovation on an existing technique.
Still, I find the complete lack of any reference to even the words "track & stack" weird, given that tracking & stacking is common practise in imaging faint asteroids. Maybe not when you use a 5-meter telescope, but with with smaller
Re: (Score:2)
Credit where credit is due.
It's called inverse synthetic aperture imaging, and people have been doing it for decades prior to 2001. The difference here is they're doing a brute-force search to discover the motion vector, rather than knowing what it is from the beginning, but then even that's not a new concept. It's merely a concept that is only recently possible due to increases in computational power.
I see a page of wtf... here (Score:1)
This is a simple and nifty idea and it works.
Good now we can find those small ez to mine rocks nearbye. Or use those rocks for other purposes... build your habitat directly in one.