Slashdot Mirror

← Back to Stories (view on slashdot.org)

Predator Outdoes Kinect At Object Recognition

Posted by timothy on from the think-well-inside-the-box dept.

mikejuk writes "A real breakthough in AI allows a simple video camera and almost any machine to track objects in its view. All you have to do is draw a box around the object you want to track and the software learns what it looks like at different angles and under different lighting conditions as it tracks it. This means no training phase — you show it the object and it tracks it. And it seems to work really well! The really good news is that the software has been released as open source so we can all try it out! This is how AI should work."

205 comments

  1. I for one.. by Dynamoo · · Score: 0

    I for one welcome our new Open Source overlords.

    --
    Never email donotemail@WeAreSpammers.com
    1. Re:I for one.. by Ruke · · Score: 4, Interesting

      I'm looking forward to looking at the GPL'd source code. There are a lot of ways to do object tracking, and they've all generally got problems, but I was rather impressed with this presentation. It was able to track the moving vehicle while it passed into and out of shadows (non-uniform saturation), as well as track that panda while it turned around (changing its shape), and it was able to distinguish a black-and-white version of the presenter's face (not based on color). It was able to recognize objects that moved off screen, which seems to indicate that it's not just drawing a snake around the moving object. Furthermore, it doesn't seem to need to be specifically programmed to track each object (as we saw the presenter just drag-and-drop a box around his hand/face.)

    2. Re:I for one.. by Anonymous Coward · · Score: 0

      Target recognized: John Connor.

    3. Re:I for one.. by Anonymous Coward · · Score: 2, Funny

      Yeah, well what would it recognise *this* as?!

    4. Re:I for one.. by LingNoi · · Score: 4, Informative

      The source code was already released. https://github.com/zk00006/OpenTLD

      There are a few more repos here.. http://www.google.co.th/#q=site:github.com+%22TLD+is+an+algorithm+for+tracking+of+unknown+objects%22&hl=en&filter=0

    5. Re:I for one.. by webmistressrachel · · Score: 2

      Warning: Goatse ahead.

      And yes, it could probably be adapted to scan links on pages being viewed in a browser for similar images to goats, and color all goatse trolls red, eliminating the need for posts like this one...

      --
      This tagline was transcoded to result in at least one smirk. If you experience failure to smirk, please consult your Gen
    6. Re:I for one.. by _0xd0ad · · Score: 1

      Amusingly enough my #a(href*=goatse.) ABP filter caught that one.

    7. Re:I for one.. by c6gunner · · Score: 1

      Warning: webmistressrachel likes to jump to conclusions without following links.

      I saw "audigoatse" and thought .... ok, this I gotta see. Turns out it's perfectly SFW.

    8. Re:I for one.. by webmistressrachel · · Score: 1

      Wow! You are of course correct, but only in this case. In fact, I think it's the first time I've ever done that!

      As you observed, I scanned the URL text quickly and looked at the mods and made a pretty foolish assumption! Thank you for the correction, I shall endevour to be more careful.

      --
      This tagline was transcoded to result in at least one smirk. If you experience failure to smirk, please consult your Gen
    9. Re:I for one.. by c6gunner · · Score: 1

      No worries. I did the same thing, except I was morbidly curious about how exactly an Audi was involved with goatse .... my imagination is doubtless NSFW.

    10. Re:I for one.. by internettoughguy · · Score: 1

      GPL, but it depends on Matlab. I wonder if anyone has got it working under Octave.

    11. Re:I for one.. by Miseph · · Score: 1

      I see you've never looked at the sticker price on an Audi if you can still muster such morbid curiosity.

      --
      Try not to take me more seriously than I take myself.
    12. Re:I for one.. by Patch86 · · Score: 1

      Could this be The Year of the Open Source UAV Attack Drone?

  2. The true test? by Anonymous Coward · · Score: 1

    The true test: can it track objects without a red dot or yellow square?

  3. Um by afidel · · Score: 0

    Kinect doesn't just track the person, it tracks the wireframe so it can figure out what the objects legs and arms are doing, that's why it has the laser rangefinder in addition to the stereo cameras.

    --
    There are 4 boxes to use in the defense of liberty: soap, ballot, jury, ammo. Use in that order. Starting now.
    1. Re:Um by SorcererX · · Score: 4, Informative

      The kinect doesn't have stereo cameras. It has one color camera which isn't really used for much, a IR projector (that projects IR dots all over the scene) and a IR camera. The IR camera uses the pixel distance between the dots to find the distance. The depth image you then get is used as input to the algorithm that detects the body parts and their orientation etc.

      --
      Any sufficiently advanced technology is indistinguishable from magic.
    2. Re:Um by am+2k · · Score: 1

      Judging from the video, it should be no problem to track individual limbs to generate a skeleton of the user. The big plus of this thing is that you don't need any special hardware at all, only a webcam is needed. Moving complexity from the hardware to the software is a big plus in the industry, because it makes the whole system much cheaper.

    3. Re:Um by marcansoft · · Score: 2

      Not the distance between dots. The camera sees exactly the same dot density regardless of depth because the projector and the camera are on the same plane (it doesn't matter if the surface is near or far, since dots will have the same angular distance when viewed from the camera). What it does measure is horizontal displacement vs. a reference image. This works because the camera and the projector are horizontally offset.

    4. Re:Um by _0xd0ad · · Score: 1

      What it does measure is horizontal displacement vs. a reference image.

      That's just another way of saying that some of the dots are displaced horizontally more than others, i.e. the horizontal distance between them is different. If the camera is positioned just to the right of the IR emitter, then two dots closer together = the rightmost dot is closer to the camera, whereas two dots farther apart = the rightmost dot is farther away.

    5. Re:Um by Blakey+Rat · · Score: 0

      Plus there's absolutely no actual comparison between this system and the Kinect at the link. And yet Slashdot is confident that it "outdoes" Kinect.

      The headline is just using Kinect as an attention-getter, when it really has very little to do with the story.

      --
      Comment of the year
    6. Re:Um by Anonymous Coward · · Score: 1

      No, the linked article made that claim.

    7. Re:Um by nedlohs · · Score: 1

      And displacement is not distance in your world, because ..?

    8. Re:Um by Blakey+Rat · · Score: 1

      Yes, BOTH Slashdot and the linked article are making the completely unsubstantiated claim.

      Me saying that Slashdot is doing it doesn't imply that the linked article isn't also. I thought Slashdot was supposed to be full of smart people, why is this basic logic so hard to figure out?

      --
      Comment of the year
    9. Re:Um by SpinyNorman · · Score: 1

      There's no reference image involved. The IR camera is offset from IR projector, and therefore sees the horizontal dot displacements caused by differences in depth.

    10. Re:Um by geekoid · · Score: 0, Troll

      You sound like you a parroting something you heard but don't really understand it.

      Stop it.

      --
      The Kruger Dunning explains most post on /. http://en.wikipedia.org/wiki/Dunning%E2%80%93Kruger_effect
    11. Re:Um by 19thNervousBreakdown · · Score: 1

      They both do things the other can't--I doubt this system does nearly as well for actual depth tracking, for instance. But, put the two together, and you would end up with something even more amazing.

      --
      <xml><I><am><so><damn>Web 2.0</damn></so></am></I></xml>
    12. Re:Um by Ksevio · · Score: 1

      The creator's website actually lists someone from MS Research as one of his advisors for the project. It's possible the Kinect and this have the same roots.

    13. Re:Um by deapbluesea · · Score: 4, Interesting

      it should be no problem to track individual limbs to generate a skeleton of the user

      I'm not so sure about that. He is using a tracking algorithm paired to a template matching algorithm. His claim is that, although both methods have high error rates, their errors are mostly orthogonal to each other. In other words, one method works better sometimes, the other method works better sometimes, and combined, they do a pretty good job. In his videos he's left out scenes where there is a large area of near constant intensity. I'm curious how his method deals with this as there aren't enough details to track, nor are there enough features to template match. Also, with arms and legs, if the texture is generally the same between the two (say you are wearing sweatpants and a sweatshirt of the same color), then there really isn't enough information for the tracker to work with in order to distinguish a leg from an arm. Straight arms and straight legs will both match the template, the tracker will likely struggle with the relatively large area of constant intensity.

      That's not to detract from Kalal's research - this is really good work - I just want to point out that it very likely suffers from a few achilles heals not mentioned in his video.

      Now pair this method with the kinect, and you might see a real improvement.

      --
      Government is not reason; it is not eloquent; it is force. Like fire, it is a dangerous servant and a fearful master.
    14. Re:Um by deapbluesea · · Score: 1

      Yes, one of his dissertation committee members is from MS's Kinect shop. I doubt his work is included in the Kinect, but it very likely will be in the next version.

      --
      Government is not reason; it is not eloquent; it is force. Like fire, it is a dangerous servant and a fearful master.
    15. Re:Um by marcansoft · · Score: 1

      Yes, dot distance does correspond to the slope of the depth, but then you'd have to integrate the resulting measurements to measure depth (that, and the dot field isn't a pattern, it's pseudorandom, so there's no trivial way to know how far apart dots are supposed to be in the first place). This isn't what the Kinect does; it directly correlates dot clusters to a reference image and measures absolute displacement against it.

    16. Re:Um by marcansoft · · Score: 1

      Displacement is distance, but (distance|displacement) between different dots in the captured image isn't the same as (distance|displacement) between a dot and its counterpart in a reference image.

      Put another way, the Kinect sees almost the same image when it looks at a wall 1m from it and a wall 5m from it (other than brightness, but that also isn't part of the algorithm because it varies depending on the material). The only difference is that one is horizontally shifted from the other. It is this absolute shift (which is compared against an internal calibration image) that is used to calculate depth. Without the reference image and the absolute comparison there would be no practical way to obtain an absolute depth measurement from an image like that.

    17. Re:Um by marcansoft · · Score: 1

      It sees horizontal dot displacement... as compared to an internal calibration reference image. There's an internally stored reference image (either as an actual image, a set of dot coordinates, or perhaps just a calibration matrix if the dot pattern is fully predictable - we don't know specifically how it's stored). If you watch a video of the IR dot image it's very easy to see how the dots move left and right as objects move near and far, but there's no way to tell the precise depth of a point in the image without a reference image to compare to.

    18. Re:Um by Anonymous Coward · · Score: 0

      In your original post you said:

      Plus there's absolutely no actual comparison between this system and the Kinect at the link.

      and then now you say:

      Me saying that Slashdot is doing it doesn't imply that the linked article isn't also.

      So in reference to your question:

      I thought Slashdot was supposed to be full of smart people, why is this basic logic so hard to figure out?

      perhaps you have not quite figured out the basic logic yourself.

    19. Re:Um by SpinyNorman · · Score: 1

      The density of dot pattern is necessarily regular (so that reflected displacements can be attributed to the surface they are projected onto). Given that it's a radial pattern (radiating outward in all directions from the source), what's fixed is essentially the angular seperation of the dots, from which some simple trig will give you the nominal seperation at any given distance from the camera, with any difference from that attributable to the angle (local depth difference) of the surface it's projected onto.

      Of course a given dot displacement could be caused by a sharp angle close to the source or a less sharp one further away, but presumably they use the average displacment withing a region to guage approximate distance, and interpret any differences accordingly.

    20. Re:Um by marcansoft · · Score: 1

      The simple trig will give you the nominal separation at the surface. Then, when you run the trig in reverse to calculate the pixel separation at the camera... you end up with exactly the same reparation regardless of depth. The camera has a radial field of view just like the projector, and pixels correspond to a fixed angle. Since the camera and the projector are on the same plane, it cancels out. Ignoring brightness (which is unreliable) and dot size (which isn't accurate enough), a flat far surface looks exactly the same as a near flat surface, except one will be horizontally shifted as a whole when compared to the other.

      Many people make this mistake, probably because they're imagining the pattern from a steep angle when viewed from outside the Kinect (or have seen YouTube videos to that effect). Our brains are also very good at associating the same angular (pixel) separation on an object that is (apparently) far away with a larger actual separation (this is why those optical illusions where two objects look different in size due to false depth cues work). However, when you view the pattern from the Kinect on the same plane as the source, and ignoring any previous knowledge of how deep parts of the image are (since, of course, finding out is the whole point), the average separation between dots is the same across the entire depth range where local depth difference is zero. Separation between dots is proportional to the derivative of the depth (local depth difference) only, but it is not proportional to absolute depth.

      For example, look at this IR dot image, as seen by the Kinect (note: half resolution, since getting the fullres raw image out via USB is problematic). If you look closely and ignore the varying dot brightness and dot size (the dots are laser beams so the virtual point source for each beam is very far behind the Kinect, and thus isn't on the same plane as the camera so dot size varies as seen by it), you'll see that the dot density is basically the same across all depths (measured in dots per area, not dot pixels per area - don't be misled by the larger dots on surfaces near the camera).

    21. Re:Um by Garridan · · Score: 1

      Imagine... a beowulf cluster of kinect predators!

      Sorry. it had to be done, for old time's sake if nothing else.

    22. Re:Um by _4rp4n3t · · Score: 1

      The creator's website actually lists someone from MS Research as one of his advisors for the project.

      No it doesn't - the MS Research guys were his examiners, not his advisers.

    23. Re:Um by exomondo · · Score: 1

      You sound like you a parroting something you heard but don't really understand it.

      Stop it.

      Yes, im sure the developer of libfreenect doesn't know what he's talking about. You, on the other hand, post an inflammatory comment with no basis.

    24. Re:Um by exomondo · · Score: 1

      No, the linked article made that claim.

      You really fail that hard at reading comprehension? Nowhere in TFA does it claim Predator outdoes Kinect at object tracking. It doesn't even mention 'object tracking' much less say it does it better than kinect.

    25. Re:Um by Anonymous Coward · · Score: 0

      no the distance between the dots will not be the same. if you take a projector, be it dots or multimedia and put it one foot away from a wall the image will be a lot smaller then if you take the same projector and put it ten feet from a wall

    26. Re:Um by IICV · · Score: 1

      Also, with arms and legs, if the texture is generally the same between the two (say you are wearing sweatpants and a sweatshirt of the same color), then there really isn't enough information for the tracker to work with in order to distinguish a leg from an arm.

      Arms have these things we like to call "hands" dangling from them, which would probably make identification of arms vs legs easier - unless you're going to go with the idea that people will use this while wearing not only clothing of a uniform color, but also gloves :)

      That's the fun thing about machine learning algorithms - they pick up on things you miss.

    27. Re:Um by Anonymous Coward · · Score: 0

      And displacement is not distance in your world, because ..?

      Because if you project at a wall going away from the camera, the distance between the dots will not change, but the whole field of dots will move toward the right (or left depending on the left/right arrangement of projector/camera)

    28. Re:Um by marcansoft · · Score: 1

      If you take a projector, be it dots or multimedia, and a camera right on top of it, the projection as seen by the camera will be exactly the same size within the camera's field of view regardless of how far away the wall is. If you move the wall back, the projection increases in size at the wall, but the wall (with the projection) decreases in size as seen from the fixed camera. The two effects cancel out.

    29. Re:Um by Anonymous Coward · · Score: 0

      Microsoft did not make the 3D technology in the Kinect, they licensed it from PrimeSense. Props are due to MS for recognizing the good tech and brining it to market in a big way.

  4. Wow, what a great idea. by Anon-Admin · · Score: 2

    1) Integrate this with a physical tracking system to move the camera to follow the target. 2) A simple program to actuate a solenoid when on target. 3) Add gun 4) train with photo 5) leave somewhere days before target arrives. 6) Profit

    1. Re:Wow, what a great idea. by Anonymous Coward · · Score: 0

      You know, that's scarily do-able.

    2. Re:Wow, what a great idea. by Kral_Blbec · · Score: 1

      While possible, it would be more complex then that. It would also have to account for wind, distance, speed, windows etc.

    3. Re:Wow, what a great idea. by Anonymous Coward · · Score: 1

      Good thing we already have algorithms for all of those.

    4. Re:Wow, what a great idea. by Anonymous Coward · · Score: 1

      Only if distance was a factor. As far as I know, Samsung already makes automated turrets. Scary indeed, but yesterday's tech. Maybe it's cheaper now.

    5. Re:Wow, what a great idea. by OzPeter · · Score: 1

      While possible, it would be more complex then that. It would also have to account for wind, distance, speed, windows etc.

      That depends on the size of the gun you use!

      --
      I am Slashdot. Are you Slashdot as well?
    6. Re:Wow, what a great idea. by royallthefourth · · Score: 1

      Oh, I'm sure someone is profiting...

      http://www.youtube.com/watch?v=xZ7FKuYPsQ0

    7. Re:Wow, what a great idea. by sammyF70 · · Score: 1

      It would also have to account for wind, distance, speed, windows etc.

      Well, it is also available for Linux and OSX,so Windows shouldn't be a problem

      --
      "DRM is like the Ford Pinto: it's a smooth ride, right up the point at which it explodes and ruins your day."-C.Doctorow
    8. Re:Wow, what a great idea. by bill_mcgonigle · · Score: 2

      Why bother with all this when a bluetooth (cell phone) listener with a range weapon is so much less complex?

      --
      My God, it's Full of Source!
      OUTSIDE_IP=$(dig +short my.ip @outsideip.net)
    9. Re:Wow, what a great idea. by Anonymous Coward · · Score: 0

      Is there anyone there? Dispensing product.

    10. Re:Wow, what a great idea. by O('_')O_Bush · · Score: 1

      http://www.lockheedmartin.com/news/press_releases/2010/052610_LM_DARPA_rifle-scope.html

      Easy.

      --
      while(1) attack(People.Sandy);
    11. Re:Wow, what a great idea. by Anonymous Coward · · Score: 0

      1) Integrate this with a physical tracking system to move the camera to follow the target.
      2) A simple program to actuate a solenoid when on target.
      3) Add gun
      4) train with photo
      5) leave somewhere days before target arrives.
      6) Profit

      I think you missed a step?!

      5.5) ???

    12. Re:Wow, what a great idea. by mr100percent · · Score: 1

      But do they have a sultry voice like the Aperture Science Turret?

    13. Re:Wow, what a great idea. by Anonymous Coward · · Score: 0

      John Connor is fucked.

    14. Re:Wow, what a great idea. by That's+What+She+Said · · Score: 1

      Damn! You just beat me! I was checking before making a Portal or Half Life reference...

    15. Re:Wow, what a great idea. by Anonymous Coward · · Score: 0

      1) Integrate this with a physical tracking system to move the camera to follow the target.
      2) A simple program to actuate a solenoid when on target.
      3) Add gun
      4) train with photo
      5) leave somewhere days before target arrives.
      6) Profit

      Now if you only combined it with a Kinect system (eyes), and other robot parts you are well on your way to 6) A Terminator

    16. Re:Wow, what a great idea. by Anonymous Coward · · Score: 0

      If by "range weapon" you mean "explosive device", you're right.

      If by "range weapon" you mean "longbow", you'll need some triangulation to actually aim the weapon.

    17. Re:Wow, what a great idea. by Locke2005 · · Score: 1

      Yeah, windows would be a real problem -- probably blue screen before the target was acquired!

      --
      I've abandoned my search for truth; now I'm just looking for some useful delusions.
    18. Re:Wow, what a great idea. by Hadlock · · Score: 1

      People have trained email spam filters to play chess. This is well within the realm of possibility.

      --
      moox. for a new generation.
    19. Re:Wow, what a great idea. by Anonymous Coward · · Score: 0

      You must be an engineer, and you solve problems. :)

    20. Re:Wow, what a great idea. by CAIMLAS · · Score: 1

      ... which is all readily available in one fashion or another. The hard work has been done; someone would just need to piece it together with this new software.

      * Temperature and humidity sensors: http://www.digitemp.com/
      * Windage: http://nslu2windsensor.sfe.se/
      * Roll it all together with ballistics calculation: http://sourceforge.net/projects/balcomp/

      That said, it's not like windage, temperature, and humidity really come into play all that much at under 200 yards. I suspect the software and servo precision at that range would be more of a contributing factor.

      The hard part, I think, would be integrating the camera with a good scope, allowing for 'automatic' zooming. (Something like this, probably: http://www.opticsbestbuy.com/ELCAN-DigitalHunter-Digital-Rifle-Scope-ELDH1.html). But with a scope like that, you'd really not need an external BDC anyway (except for humidity and wind, I suppose). Still, even without humidity or wind, something like this would give an inexperienced but steady shooter twice, maybe 3 times the effective range. I'm sure the same would go for a 'terminator' robot.

      Then, you'd need to work out some rudimentary logic for scanning, polling, etc. of the scope so your field of view is large enough to actually detect the target, and I suspect you'd be 'golden'. (A better setup would probably be to have two, independent but potentially syncronizable sensors next to each other; one would be for scanning (ie you'd not notice it moving) and the other would have the rifle, which, upon target recognition, would use the 'sensor' data to lock in the trajectory.)

      Anyway, like I said: the hard work has been done now. Building the device would be relatively easy and someone with, say, a background in a technical field (a competent engineer type) could piece it together in a handful of dedicated weekends, I suspect.

      --
      ~/ssh slashdot.org ssh: connect to host slashdot.org port 22: too many beers
    21. Re:Wow, what a great idea. by Anonymous Coward · · Score: 0

      Murder by robot camera is no scarier than normal murder, the implication that open-sourcing this has somehow created means where means did not exist before is asinine.

    22. Re:Wow, what a great idea. by cgenman · · Score: 1

      We all know your automatic evil guns of death should be written on QNX. But after the corporate committee is done with it, it will probably just be a Silverlight plug-in.

      --
      The ______ Agenda
    23. Re:Wow, what a great idea. by RooftopActivity · · Score: 0

      Sorry, I can't come to work this weekend, my weather station is trying to attack me.

  5. Yo Dawg by xMrFishx · · Score: 1

    I heard you like you like Predators, so we put a Predator on your Predator so you can spy while you spy.

    Watched TFV on TFA, very interesting. Something to play with soon I think.

    1. Re:Yo Dawg by netdigger · · Score: 0

      Predation?

  6. Predator UAV? by Singularity42 · · Score: 0

    I'm assuming the robot plane can track objects pretty well before it disconfigures them.

  7. State of the art? Yes. Breakthrough? No. by Anonymous Coward · · Score: 0

    This method uses fairly standard techniques (tracking and on-line learning), and puts them together nicely. It is very nice work, but hardly a breakthrough in AI.

    Also, this has nothing to do with Kinect. This tracker uses a 2D camera to track 2D image patches, while learning their shape. The Kinect is a 3D sensor which is used for tracking *articulated* models, such as people.

    1. Re:State of the art? Yes. Breakthrough? No. by alexborges · · Score: 0

      "The Kinect is a 3D sensor which is used for tracking *articulated* models, such as people."

      Big words for a fucking distance sensor.

      --
      NO SIG
    2. Re:State of the art? Yes. Breakthrough? No. by countertrolling · · Score: 2

      It's a breakthrough in the price of AI..

      --
      For justice, we must go to Don Corleone
  8. yeah by sentientbeing · · Score: 1

    Yeah thats how it starts off. First you're like "ooh ahh look at the cute little robot isn't he pretty walking around by himself" then later theres running and screaming and then its all like "Newsflash :Bombing in Midtown, USA - Cyborg liberation front demands equal rights for robots".

    --

    ------
    beware he who would deny you access to information, for in his mind he dreams himself your master
    1. Re:yeah by hellkyng · · Score: 1

      No kidding and then the debate really heats up:

      1. Robots want to be able to marry > Marriage is between a fleshing and a fleshing (cyborgs or flesh covered robots allowed too in Massachusetts)
      2. FemBots want to be able to choose to have an EMP burst > EMPs are nuclear based malicious malfunctions!
      3. Robots want to "open-source" themselves, no debate ensues but its only legal in the outskirts around Las Vegas.

      Won't someone think of the child-bots?

    2. Re:yeah by Anonymous Coward · · Score: 0

      The climate's easy on my switches and my stable's full of bitches.

      I got hoes on Las Vegas Boulevard and my software is always hard.

    3. Re:yeah by Conspiracy_Of_Doves · · Score: 1

      I think this might be relevant.

      And then, eventually we will need this

      --

      Technoli
    4. Re:yeah by eriqk · · Score: 1

      Yeah thats how it starts off. First you're like "ooh ahh look at the cute little robot isn't he pretty walking around by himself" .

      We were warned.

  9. with a name like that, by Anonymous Coward · · Score: 0

    what could possibl

  10. sky net is growing by Anonymous Coward · · Score: 0

    sky net is growing

  11. Very nice. by Animats · · Score: 2

    Very nice.

    There are other systems which do this, though. This looks like an improvement on the LK tracker in OpenCV.

    This could be used to handle focus follow in video cameras. Many newer video cameras recognize faces as focus targets, but don't stay locked onto the same face. A better lock-on mechanism would help.

    1. Re:Very nice. by Anonymous Coward · · Score: 0

      He combines Lucas-Kanade with a classifier.
      That's a classic predictor-corrector approach, but there are some nice ideas added like online learning and his N-P classifier.
      Though it's not clear if it works in ill-conditioned settings which mess up his classifiers.
      But very nice work nonetheless.

  12. Neat by symes · · Score: 1

    That was a very nice demonstration and well done to Zdenek Kalal. That said, there's a bunch of trackers out there and what I find is that none of them do well in a noisey environment where there's a bunch of similar items. Security cameras have to work in the rain, snow, fog, low light conditions. So Zdenek, if you are listening, how real-word can you go with this?

    1. Re:Neat by Amouth · · Score: 1

      towards the end there is an example of it tracking a car on the freeway - i think that might fit the bill

      --
      '...if only "Jumping to a Conclusion" was an event in the Olympics.'
    2. Re:Neat by fractoid · · Score: 1

      The face tracking bit had it tracking his face for a while, once it'd learned enough example images he held up what looked like a class list sheet with a bunch of small black-and-white mug shots on it, and it picked his face out. That seemed like a fair test for rejecting false positives (it was actually looking for HIS face, not just A face).

      --
      Rampant carbon sequestration destroyed the Dinosaurs' tropical paradise. I'm here to help repair the damage.
    3. Re:Neat by Anonymous Coward · · Score: 0

      As someone who studied the SIFT transform for my research thesis, here are my findings:

      The SIFT algorithm works well in conditions where there is little background noise or few similar objects in an image and when the target object is sufficiently complex.

      Background noise causes issues as it results in a large number of SIFT descriptors being output which then need to be filtered out during matching (Hough transform). As the Hough transform has a generally high order of complexity it can be slow to compute and a large number of descriptors increases the chances of false positive outputs significantly. This extent of this depends of your thresholding algorithm but in general holds true.

      When the target object is not sufficiently complex by which I mean has few distinctive edges, like a logo, then the number of descriptors which describe that object can be relatively small, though adjusting the threshold of the SIFT algorithm can help. This is obviously an issue as if only a small number of descriptors can be matched against, then chances of finding a good matching set from an image are reduced when the image is warped or obscured.

      This implementation does have advantages which will help the SIFT perform better. By identifying where the object you are interested in is in the start of a sequence allows the original position of the object to be taken into account and so can improve your algorithm by using this in the final decision making. Successive frames can then use the position identified in the previous and so help tracking.

      Nice demo.

  13. Why still fooling with ONE camera? by ackthpt · · Score: 2

    Shouldn't we be developing AI to use two? I mean, we have two eyes (most of us, condolences to those who do not, no disrespect intended) and we recognize objects, dept of field and rates of change within three dimensions, using them.

    --

    A feeling of having made the same mistake before: Deja Foobar
    1. Re:Why still fooling with ONE camera? by Anonymous Coward · · Score: 1

      maybe we should give the AI an appendix too...

    2. Re:Why still fooling with ONE camera? by Anonymous Coward · · Score: 0

      Shouldn't we be developing AI to use two? I mean, we have two eyes (most of us, condolences to those who do not, no disrespect intended) and we recognize objects, dept of field and rates of change within three dimensions, using them.

      But them the robots will be just as fooled by stereoscopic 3D as we are!

    3. Re:Why still fooling with ONE camera? by Anonymous Coward · · Score: 0

      Because you don't need it to do a pretty good job and it takes vastly less calibration to use.

      Close one eye. Now track something. If you can do it, why shouldn't our AI be able to?

    4. Re:Why still fooling with ONE camera? by Ruke · · Score: 2

      The only thing two cameras really nets you is more reliable depth perception; however, this requires regular calibration, as minute shifts in cameras (say, from being jostled around while moving) can translate to large errors if your focal points aren't exactly where you think they are. It's often easier to track movement using the change-in-size of your object, and have a separate specialized depth-sensor (sonar, laser, etc) to perform depth measurements when you need them to be exact.

    5. Re:Why still fooling with ONE camera? by SorcererX · · Score: 1

      Indeed, I had a project a while ago in which we used two separate cameras mounted on a robot. We barely had to touch the camera rig to get the cameras out of alignment. Once it was mounted it was pretty good though. There are companies such as Point Grey Research that makes Stereo Vision Camera setups that stay aligned properly even if you move the rig around, but such setups aren't particulary cheap.

      --
      Any sufficiently advanced technology is indistinguishable from magic.
    6. Re:Why still fooling with ONE camera? by smelch · · Score: 2

      And a glossary.

      --
      If I can just reach out with my words and touch a butthole, just one, it will all be worth it.
    7. Re:Why still fooling with ONE camera? by Abstrackt · · Score: 1

      Shouldn't we be developing AI to use two?

      Why? One camera is cheaper to purchase and maintain than two and this software seems to do just fine with one.

      --
      They say a little knowledge is a dangerous thing, but it's not one half so bad as a lot of ignorance. - Terry Pratchett
    8. Re:Why still fooling with ONE camera? by theY4Kman · · Score: 1

      You are surely disrespecting those with one eye by suggesting they can't recognize objects.

    9. Re:Why still fooling with ONE camera? by JayJay.br · · Score: 1

      And an index. What the hell, give 'em all five fingers.

    10. Re:Why still fooling with ONE camera? by Anonymous Coward · · Score: 0

      Mod parent up for teh lulz!

    11. Re:Why still fooling with ONE camera? by Conspiracy_Of_Doves · · Score: 1

      Why stop at 2?

      --

      Technoli
    12. Re:Why still fooling with ONE camera? by tirerim · · Score: 1

      We really only use our binocular vision for depth perception at fairly small distances (less than 10 meters); more than that, and we're just relying on things like relative size, perspective, and motion parallax (which assist at smaller distances, too). If we're designing robot surgeons or something else that needs equally fine sensitivity, then yes two cameras would be the way to go, but for most purposes they're just unnecessary.

    13. Re:Why still fooling with ONE camera? by ElectricTurtle · · Score: 1

      In Greece that's more offensive than just one anyway.

      --
      I support the Slashcott and will not be reading or commenting from 2/10/14 to 2/17/14. Beta is steaming pile of dog shit
    14. Re:Why still fooling with ONE camera? by Anonymous Coward · · Score: 0

      Multiple cameras should dramatically improve this system though.
      What helps it to do better is the time spent looking at different angles of the object.
      What makes it do worse is if the angle of the object shown is suddenly too different from what has previously been observed.
      For this, having a fly's eye camera should dramatically improve speed of recognition and reduce incidence of no recognition.

    15. Re:Why still fooling with ONE camera? by durrr · · Score: 1

      And you're better off saying blind people can't recognize objects? Just wait until they get their hands on you.

    16. Re:Why still fooling with ONE camera? by Hadlock · · Score: 1

      4 billion years of evolution, and 99% of living creatures have a pair of eyes. Even flies, with compound eyes, have a pair of them. There seems to be something useful - such as a wider field of view - to having two, rather than one. Humans and most primates have stereoscopic vision, but that's a relatively rare event in nature.

      --
      moox. for a new generation.
    17. Re:Why still fooling with ONE camera? by HeckRuler · · Score: 1

      You're clinging to the piano top in the sea of ignorance. The human form (or even the mammalian one) is not the best to emulate.

    18. Re:Why still fooling with ONE camera? by JanneM · · Score: 2

      "99% of living creatures have a pair of eyes."

      Most of those eyes - flies included - are not used for stereoscopic perception. They have two eyes because one eye typically covers less than half the visual field. Most animals' eyes are pointed away from each other, with very little or no visual overlap anywhere.

      Depth perception mostly does not need stereoscopy. If it did, one-eyed people would hardly be able to walk or feed themselves, never mind drive a car or other things.

      Stereovision is good mainly for precise location within, oh, a few meters, in the case of primates. The kind of distance and precision you'd need to move between tree branches. For longer distances, or for less accuracy, other cues are sufficient.

      --
      Trust the Computer. The Computer is your friend.
    19. Re:Why still fooling with ONE camera? by Skidborg · · Score: 1

      So I'm curious. How do humans calibrate their depth perception?

      --
      Supporter of the +1 Over Dramatic mod option. In memory of apk.
    20. Re:Why still fooling with ONE camera? by Skidborg · · Score: 1

      Okay, let's go with fish... two eyes. Birds... still two. Insects? Reptiles? Spiders? What only has one eye?

      True, some of them don't use them for stereo vision, but so far as natural selection is concerned, two or more eyes seems to be the winner if you're going to have eyes at all.

      --
      Supporter of the +1 Over Dramatic mod option. In memory of apk.
    21. Re:Why still fooling with ONE camera? by Hadlock · · Score: 1

      Did you want to Drop Some Knowledge on slashdot, or did you not read the last sentence in full? What part of "relatively rare" was unclear?

      --
      moox. for a new generation.
    22. Re:Why still fooling with ONE camera? by Anonymous Coward · · Score: 0

      Why? One camera is cheaper to purchase and maintain than two and this software seems to do just fine with one.

      Duhhhh... so robots don't have to rely on IR pattern projection?

    23. Re:Why still fooling with ONE camera? by the+biologist · · Score: 1

      We move around and grab at things. Our eyes occasionally return the wrong distance, but upon moving/etc the problem is realized.

    24. Re:Why still fooling with ONE camera? by tixxit · · Score: 1

      Lots of robots do use 2 eyes (cameras) for 3D vision. http://opencv.willowgarage.com/documentation/camera_calibration_and_3d_reconstruction.html

  14. What should I name my video tracking technology? by asdbffg · · Score: 1

    Oh... I know. "Predator." That's not a loaded, terrifying term at all.

  15. Not a breakthrough by Dachannien · · Score: 4, Informative

    This isn't a breakthrough. Much of the technology for tracking objects in this way has been out for about a decade. See this Wikipedia article for one technique for doing this:

    http://en.wikipedia.org/wiki/Scale-invariant_feature_transform

    1. Re:Not a breakthrough by immakiku · · Score: 2

      I think the breakthrough is the speed improvements to do this in real time on reasonable commodity hardware?

    2. Re:Not a breakthrough by bughunter · · Score: 3, Funny

      Indeed. I've worked on some military programs that track and intercept, umm... things... for various purposes... that use this very same image-based tracking algorithm. But instead of painting a red dot or drawing trails, it steers a, umm... vehicle... that... uh... delivers candy.

      Yea. Candy.

      Euphemism aside, he's done a very nice job of integrating it with commercial hardware and software. It's still impressive.

      --
      I can see the fnords!
    3. Re:Not a breakthrough by durrr · · Score: 3, Insightful

      Unless you're a ballistic missile or insurgent you're likely to never see anything of those military systems. If we invent a matter replicator and only use it for creating delicious topping for ice cream it still wouldn't be as much of a waste as the military hush-hush superadvanced fancypants-never-for-good-use systems.

    4. Re:Not a breakthrough by Anonymous Coward · · Score: 0

      Yup. I've seen demos and worked with OpenCV code that did similar tracking using Normalized Cross-Correlation for Template Matching, which are rather basic methods, especially compared to SIFT. They worked pretty well on my 2008 MacBook for tracking a "template" (basically whatever is in a rectangle that you select) in a real-time video stream from the webcam.

      Here's a link to and a video of the demo project... You can see how well even such basic methods perform, and where they fall apart:
      http://sourceforge.net/projects/jmatchtemplate/
      http://www.youtube.com/watch?v=TrAM8_Fw3yQ

      I'm sure this dude's done something pretty cool, but the object tracking in his videos is pretty old hat. I guess the real novelty will be detailed in the papers.

    5. Re:Not a breakthrough by lennier · · Score: 1

      If we invent a matter replicator and only use it for creating delicious topping for ice cream

      I see you've been visiting Heston Blumenthal.

      --
      You are not a brain: http://books.google.com/books?id=2oV61CeDx-YC
    6. Re:Not a breakthrough by CrustaceanSoup · · Score: 1

      See SURF. And there's real-time GPU implementations of SURF and SIFT that run on the order of 60 FPS or faster.

    7. Re:Not a breakthrough by Anonymous Coward · · Score: 0

      After trying the code out, I think what's unique about TLD is that it combines feature search with optical flow using the Lucas-Kanade algorithm. This allows the algorithm to create a library of feature matches and ultimately weight the output of the two techniques to come up with a final target position.

    8. Re:Not a breakthrough by Anonymous Coward · · Score: 0

      Yep the military does sure make all those useless things, like radar and microwaves and rocket engines and jet engines and nuclear fission and gps and arpanet. Too bad military R&D never results in any civilian applications, such a waste.

    9. Re:Not a breakthrough by Anonymous Coward · · Score: 0

      I suspect every rebel opposing Qadaffi in Libya and every woman (or insufficiently fundamentalist man) who suffered under the Afghani Taliban and every Kuwaiti who prefers their non-homocidal tyrants to Saddam Hussein and every Kurd and many Bosnians might disagree with that blanket "never-for-good-use" assessment of yours. So, too, might the people of South Korea and Taiwan and, arguably, Japan and farther, who breathe free air without us firing shots, largely due to the deterence of our force of arms.

      I don't agree that all of those military actions were the best things to do --for America, or for the world-- or that our current levels of military spending make any kind of goddamned sense. But we are, basically, a force for good in the world. Net, anyway. It's not even close to neutral.

    10. Re:Not a breakthrough by Anonymous Coward · · Score: 0

      The military and serial killers have at least one thing in common, they refer to their victims as "things".

    11. Re:Not a breakthrough by gl4ss · · Score: 1

      so the breakthrough is .. cheap fast pc's? that's not a breakthrough. seems like nice code though, looking for an application(for example, the problem with a minority report ui isn't actually that it's hard to do, it's that it's hard to imagine any pc work where it would be the way to go).

      --
      world was created 5 seconds before this post as it is.
    12. Re:Not a breakthrough by ceoyoyo · · Score: 1

      He's doing object tracking in realtime without using a GPU. And it's learning and improving it's performance while doing that tracking, not being trained ahead of time like most SURF/SIFT implementations.

    13. Re:Not a breakthrough by CrustaceanSoup · · Score: 1

      SURF will run in realtime on a CPU (not at 60+ FPS on most commodity CPUs, but still acceptably), but yeah, his thing is definitely not that. For feature detection it uses something like haar-like detectors like SURF does (hence the speed), but there's a lot more to it. The papers are worth reading, it's a really neat technique.

  16. Matlab by Anonymous Coward · · Score: 0

    This runs inside of, or is somehow dependent on Matlab. I know little about Matlab, but I find that somewhat odd that it cannot be implemented as a standalone application. What magic of Matlab cannot be easily reproduced to make this a standalone app?

    1. Re:Matlab by JonySuede · · Score: 1

      tons and tons of exotics FORTRAN like code that is shared from grad student to grad student

      --
      Jehovah be praised, Oracle was not selected
    2. Re:Matlab by _0xd0ad · · Score: 1

      The "magic" is that Matlab has a lot of very fast and powerful built-in matrix operations. Can be reproduced, yes. Easily, no.

      However, Matlab also has a compiler package that creates stand-alone executable files...

    3. Re:Matlab by Anonymous Coward · · Score: 0

      There's nothing special about Matlab and the performance of its matrix operations. It's just a nice development environment and he chose to use it, that's all. I bet in a few weeks there'll be a matlab-free reimplementation.

    4. Re:Matlab by _0xd0ad · · Score: 1

      Sure. All you'd have to do is use the compiler package to produce an executable and then decompile it into whatever language you prefer. Pretty it up some and call it good. Like I said, nothing "magic" about Matlab, just a bunch of useful built-in functions that he chose not to re-invent.

    5. Re:Matlab by norminator · · Score: 1

      The standalone executables still depend on a runtime environment that has to be installed on the target computer, so you still can't just drop the .exe on any old computer and run it.

    6. Re:Matlab by _0xd0ad · · Score: 1

      My recollection was that you could, but maybe it was a different compiler package than the one you're familiar with. Or maybe I'm remembering wrong.

  17. Great News! by Anonymous Coward · · Score: 0

    Ok guys, seriously. This isn't good news. This is just one more step towards the inevitable and eternal oppression of the majority of humanity through automation.

  18. Great video, but by AnonymmousCoward · · Score: 0

    @1:40 "Usefull for disabeled people"...

    But apparently not useful for PhD students who can't spell

    1. Re:Great video, but by hotkey · · Score: 5, Informative

      who can't spell

      I guess you're fluent in Czech?

    2. Re:Great video, but by AnonymmousCoward · · Score: 1

      Really? Your comment was modded up while mine was modded down...
      I was being ironic. Obviously over the head of most /.'ers (cue eyeroll)

    3. Re:Great video, but by Anonymous Coward · · Score: 0

      It was still an insult to his English (which is his 2nd language). We have enough grammar nazis here, no offense.

  19. Matlab required? by cmpalmer · · Score: 1

    How usefully open-source can it be with a commercial library requirement?

    --
    -- stream of did I lock the front door consciousness
    1. Re:Matlab required? by royallthefourth · · Score: 1

      It's not that bad; the code can be ported to a useful language and distributed. It's an extra step but it's far from worthless (as far as software goes).

    2. Re:Matlab required? by Anonymous Coward · · Score: 0

      Apparently, people have been able to run it with GNU Octave with minor changes. follow the github project and the discussion group if that really interests you.

    3. Re:Matlab required? by Anonymous Coward · · Score: 0

      Open Source and Free Software are two different things.

    4. Re:Matlab required? by Anonymous Coward · · Score: 0

      Not as much difference from reading the related papers and writing from scratch. It might even be easier than deciphering the ugly matlab syntax. I don't blame the author, that's how most PhDs end up anyways, an ugly pile of code that needs to rewritten from scratch to actually be useful. Someone will most likely re-implement it from scratch, and his original code will probably wither and die in obscurity.

    5. Re:Matlab required? by ceoyoyo · · Score: 1

      I've found it's considerably easier to implement something described in a paper (most people's algorithm papers are TERRIBLE) when you've got a reference implementation, even if it is ugly MatLab code, to work from.

      A nice C implementation would be fantastic though. Maybe someone will merge it back into OpenCV.

  20. What if... by Anonymous Coward · · Score: 0

    You point it at a hall of mirrors?

  21. January by Taur0 · · Score: 1

    This video was uploaded in January, and it's on slashdot NOW?

    1. Re:January by amliebsch · · Score: 1

      I'm guessing somebody was annoyed by the Kinect stories recently...

      --
      If you don't know where you are going, you will wind up somewhere else.
    2. Re:January by Anonymous Coward · · Score: 0

      Actually, it was to be uploaded in January, but the Chinese government suppressed it till today, due to their new time travel restrictions...

    3. Re:January by kevinNCSU · · Score: 1

      This video was uploaded in January, and it's on slashdot NOW?

      You mean people didn't IMMEDIATELY take notice of a video posted by a Czech student for his phd thesis? Good God, who's manning the internet?!?!

  22. Re:What should I name my video tracking technology by PReDiToR · · Score: 2

    Bloody stupid name if you ask me ...

    I mean they even spelled it properly and everything.

    --

    Do not meddle in the affairs of geeks for they are subtle and quick to anger
  23. Totally different things. by tgd · · Score: 1

    Kinect is how you feed data to an image recognition/tracking algorithm, Predator is that algorithm. The software side of Kinect has support for efficiently tracking items, but that is so you have the most CPU left for a game. That was the trade-off.

    Kinect hardware can do something very useful that Predator can't -- it can tell how far away something else (and thus, judge position or size more accurately).

    The predator algorithim (and other ones no doubt under development) using the two sets of data from a Kinect camera will still be superior to an algorithm using just one set of data.

    1. Re:Totally different things. by darien.train · · Score: 1

      The predator algorithim (and other ones no doubt under development) using the two sets of data from a Kinect camera will still be superior to an algorithm using just one set of data.

      This is what I'm thinking as well. I've done a bit of Kinect data stream/parsing experiments with other input types (like adding a touch screen to record "impact" data while the kinect detects telemetry) and I think adding predator will be pretty damn useful.

      I can't really go into the really killer kinect tracking shit I've been working on (NDA) but predator might solve a few issues I've been having.

      Exciting!

      --
      I don't know how many years on this Earth I got left. I'm going to get real weird with it. - Frank Reynolds
  24. No training phase? by MosnarSgnik · · Score: 1

    How is showing it the object at different angles and different lighting conditions not a training phase?

    1. Re:No training phase? by Anonymous Coward · · Score: 0

      It is given an initial patch to track. It learns how it changes over time and updates the tracking template.

    2. Re:No training phase? by ifrag · · Score: 2

      I think what was meant is no independent training phase. The training is in parallel with actual use.

      --
      Fear is the mind killer.
  25. Nothing new or great by Anonymous Coward · · Score: 2, Interesting

    As a person who does on a daily to daily basis research on object tracking, and having seen implementations and performances of many trackers (including this one) on real world problems (including gaming), this is nowhere a new approach or an approach which outperforms many other ones published in recent computer vision conferences.

    From TFA:
    "It is true that it isn't a complete body tracker, but extending it do this shouldn't be difficult."

    Going from this to body tracking is a HUGE step, it's not a really easy thing to do. I don't know there is a strange hype around this one which I can't really understand the reason, it's coming up on many websites etc, while as I said not being a great tracker.

  26. Re:What should I name my video tracking technology by Anonymous Coward · · Score: 0

    czechs, being generally weak and defenseless, have a jealousy hard-on for the american military system

  27. I'm not surprised by tooslickvan · · Score: 1

    Predator outdoes everything and everyone except for Arnold Schwarzenegger.

  28. Re:What should I name my video tracking technology by Anonymous Coward · · Score: 0

    "Moley Mole" was the first name that came to me when watching the video.

  29. Kirk by Anonymous Coward · · Score: 0

    Beam me up, Scotty.

  30. Not open source by interiot · · Score: 1
    It may not really be open source. The author says it's available under the GPLv3. But the author also says something completely contradictory:

    A license has to be purchased for using TLD in a commercial project. The licencing is managed by the IP owner, the University of Surrey and the fee is subject to negotiation.

    1. Re:Not open source by pmontra · · Score: 1

      It means that if you want to use that code in a closed source program you can do it if you buy a license from him. He owns the copyright so he can multiple-license the code. GPL doesn't prevent that and it is quite common.

    2. Re:Not open source by kelemvor4 · · Score: 1

      According to the page you linked:

      Free Version

      TLD can be downloaded for testing in a chosen application. We provide a precompiled demo (Windows) and a source code that is released under GPL version 3.0. In short, it means that any distributed project that includes or links any portion of TLD source code has to be released with the source code under the GPL version 3.0 license or later.

      Commercial Version

      A license has to be purchased for using TLD in a commercial project. The licencing is managed by the IP owner, the University of Surrey and the fee is subject to negotiation. Please contact the University of Surrey for further information.

    3. Re:Not open source by Anonymous Coward · · Score: 0

      What he probably means is that if you wish to use the code without having to adhere to the GPLv3 restrictions, then he would be willing to sell you a license. There's nothing wrong with that from an open source perspective.

    4. Re:Not open source by Anonymous Coward · · Score: 0

      They might mean it is dual-licensed i.e. available under GLPv3 if you are willing to keep the source open or available under a commercial license if you want to keep your modifications closed.

    5. Re:Not open source by nedlohs · · Score: 1

      It's not contradictory, it's just incorrect terminology.

      By "commercial" they mean "non-GPLv3 compliant", which is wrong since GPLv3 licenses can be used for commercial products just fine. And you could not want to use the GPLv3 for a non-commercial project...

      But it's a common error, since the overlap is rather large.

    6. Re:Not open source by Em+Adespoton · · Score: 1

      ClosedTLD is licensed under a commercial license; OpenTLD is under GPLv3. Eventually these will likely fork (if anyone else adds code to the GPL'd version).

      More info (and code) for OpenTLD here:
      http://groups.google.com/group/opentld
      https://github.com/zk00006/OpenTLD

    7. Re:Not open source by LingNoi · · Score: 1

      The author originally released it as open source under the GPL but then withdrew the link from his site when he realised how much attention it was getting. Some people have released the original GPL copy on github for others to use and distribute.

    8. Re:Not open source by Em+Adespoton · · Score: 1

      Oh, and a fork: https://github.com/pascalbertrand/OpenTLD/blob/osx-octave/ that supports OS X and uses Octave (open source) instead of Matlab. This means that the entire toolchain is now OSS, and compiles on OS X!

  31. "This is how AI should work"? by Provocateur · · Score: 1

    Now why did you have to go and say that? Don't you know they hate it when you tell them what they're supposed to do?

    Wouldn't be surprised if the robot uprising took place tonight. At least, I know who pushed them over the edge.

    --
    WARNING: Smartphones have side effects--most of them undocumented.
  32. I had a Terminator joke... by hoppo · · Score: 1

    but I'm just going to take it to Fark.

    1. Re:I had a Terminator joke... by lennier · · Score: 1

      but I'm just going to take it to Fark.

      Take it to this guy.

      To be or not to be? (gunshots) Not to be.

      --
      You are not a brain: http://books.google.com/books?id=2oV61CeDx-YC
  33. Actually... this isn't how A.I should work! by Tasha26 · · Score: 1

    Giving an initial starting point and/or tracing the object amounts to pure cheating. This is going backwards in time in terms of evolution of A.I. The object to track (a scale/rot/trans-invariant shapes database) and its intial starting point on the picture is a big problem in Computer Vision. This software ain't a breakthrough if the computer can't track a known object by itself.

    1. Re:Actually... this isn't how A.I should work! by Anonymous Coward · · Score: 0

      Cheating is, I think, a rather harsh term. When the object is first marked, is isn't "known", it's being learnt at the same time. Based on the video, in a very short amount of time (concurrent learning and tracking), it can continue to track the object even if it has left the scene and returned, so without having an initial point in the picture. So either you're saying you want it to be able to track an object from scratch, without being given any information on the object at all (which is rather silly), or you mean it should be doing exactly what it is doing, and just haven't bothered to watch the demonstration fully.

    2. Re:Actually... this isn't how A.I should work! by Tasha26 · · Score: 1

      One might start by scanning every pixel from location (0,0) and try to match "a" shape or maybe to speedup, subdivide the picture and for each child run a search? Just kidding, there are primitive methods (Computer Vision 101) to discover a "rough" initial starting point. On another note, the Darpa A.I car which drove itself through the desert was not given any such human input. There's a great PBS/Nova video on it.

  34. Um...err... NO!!!! by SpinyNorman · · Score: 5, Informative

    No - wrong on all counts.

    - Kinect doesn't have stereo cameras (it has an IR camera for depth perception and a visible light camera for other usage)

    - Kinect doesn't use the visible light camera for body recognition. Recognition is based on the depth map provided by the IR grid projector and IR camera.

    - Kinect doesn't operate like a laser rangefinder (it operates via structured light displacements, not via light pulse reflection times)

    - Kinect doesn't track a wireframe (it tracks independent body parts)

    How you got modded as "4 - informative" is beyond me. The blind leading the blind.

    The way Kinect works is by projecting a dense evenly-spaced grid of IR dots (i.e. structured light) on the scene, then using it's IR camera (horizontally offset from the grid projector) to pick up the reflected dot pattern.

    Due to depth differences in the scene, and the offset of the IR camera from the IR projector, the reflected dot pattern is not evenly spaced - the dots are horizontally displaced based on depth. To understand this, consider shining two parallel beams of light at a) a flat surface, and b) a surface angled at 45 degrees away from the light source. If you took a step sideways away from the light beams and looked at their reflections of the two objects, the dot (beam) separation on the flat surface would be the same as the true beam separation, but the dot separation on the angled surface would be increased. by an amount you could calculate using simple trig.

    In order to operate in real-time with low cost, a dedicated chip processes the IR camera image and converts the dot displacements into the corresponding depth map.

    The clever, and somewhat counter-intuitive, part is how Kinect then turns this depth map into a body part map. The basic idea is that it probabilistically maps local clusters of depths to body parts (via having been trained on a huge manually body-part-labelled image set), then converts these local probabilities into larger scale body part labels (i.e. if 60% of the local clusters in a region say "hand" , then the region is labelled as a hand). This way it doesn't track overall body postion or a wireframe, but rather independently tracks body parts (which is why it has no trouble correctly tracking muliple partially occluded people in frame).

    1. Re:Um...err... NO!!!! by Anonymous Coward · · Score: 0

      > The blind leading the blind.

      I thought that was the motto of Slashdot... and indeed of our society as a whole.

      (Protip: If you don't know at least quantum physics and are an expert in social dynamics / neuropsychology, you're blind to me.)

    2. Re:Um...err... NO!!!! by Skidborg · · Score: 2

      Because quantum physics and neuropsychology make you an expert on object tracking AI and any other field? What if specializing has made you blind to things outside of your field of expertise? You might be making a valid point somehow, but your reeking aura of arrogance makes you a nuisance rather than a good leader.

      --
      Supporter of the +1 Over Dramatic mod option. In memory of apk.
    3. Re:Um...err... NO!!!! by initialE · · Score: 1

      What bugs me is that with the size and shape of the Kinect, why doesn't it use stereo cameras? A lot of design work may have gone into the thing, but it seems like a large amount of the things in it are just redundant. Extra wide wingtips. A robotic arm with only 1 degree of freedom. No mount points. I didn't take mine apart, but I'm guessing there is a lot of open air inside that casing.

      --
      Starbucks, Harbuckle of Breath.
    4. Re:Um...err... NO!!!! by Anonymous Coward · · Score: 0

      Are experts on social dynamics generally douchey enough to think using "protip" makes them seem smarter?

    5. Re:Um...err... NO!!!! by gnapster · · Score: 1

      What bugs me is that with the size and shape of the Kinect, why doesn't it use stereo cameras?

      Because stereo cameras would not help with object detection as currently implemented.

  35. In Soviet Russia... by Kamiza+Ikioi · · Score: 2

    ...the software learns you!

    --
    I8-D
  36. It's not Goatse, it's a genuine ad for Audi cars! by Anonymous Coward · · Score: 1

    Warning: Goatse ahead.

    Er, but did you actually look at the image? It's not goatse itself, it's an advert for Audio cars that's appeared on public billboards throughout the UK(!)

    Spotted the similarity myself instantly, but having checked online it appears that lots of other people in the UK spotted and uploaded the same poster as well. Not surprised, more surprising that someone at the ad agency didn't spot it!

  37. Check off another box on the Cyberdyne list! by Anonymous Coward · · Score: 0

    Ambulatory across a variety of terrain - CHECK
    Wireless encrypted connectivity to a central hub - CHECK
    Interchangeable weapons - CHECK
    Perception of humanity as threat - STILL WAITING ON AI (but won't be long after)
    Advanced tracking - CHECK

    Get your popcorn ready, should be a helluva show!

    1. Re:Check off another box on the Cyberdyne list! by Killjoy_NL · · Score: 1

      Perception of humanity as threat - STILL WAITING ON AI (but won't be long after)

      Nice timer here :) http://www.aperturescience.com/a/b/c/d/g/h/abcdgh/

      --
      This is the sig that says NI (again)
  38. Re:It's not Goatse, it's a genuine ad for Audi car by webmistressrachel · · Score: 1

    Wow! You are of course correct, I have been trolled! I feel pretty stupid now.. the post was probably crafted with that URL to catch people out!

    Sorry folks... post redacted!

    --
    This tagline was transcoded to result in at least one smirk. If you experience failure to smirk, please consult your Gen
  39. Summary is wrong by Anonymous Coward · · Score: 0

    "All you have to do is draw a box around the object you want to track and the software learns what it looks like at different angles and under different lighting conditions as it tracks it. This means no training phase — you show it the object and it tracks it."

    Um, what?

    Manually labeling something and giving the software the opportunity to learn the pattern in the box... that right there is training. The software is trained on one set of pixels so it can continue to detect similar groupings of pixels in future image frames.

    Someone needs to go study their old Pattern Recognition 101 notes...

  40. This is not AI... by Anonymous Coward · · Score: 0

    These "neural networks" and other algorithms to teach recognition to computers aren't AI. Worse: it isn't bringing us any closer to AI than what we had 20 years ago... Or the definition of "AI" is a very, very, sad one.

  41. Predator to DMX? Any Ideas? by t0qer · · Score: 1

    I work at a karaoke bar. http://www.justin.tv/7bamboo I'd really like to use this to track singers as they move around the room, or have spotlights follow.

  42. Over hyped? by Anonymous Coward · · Score: 0

    IAMA Computer vision and robotics researcher and I find the Predator video very over hyped. This is nothing new.

    As mentioned by previous posts, tracking with ongoing model updates isn't new. It has been around pretty much since the information filter (Kalman, condensation etc) days. As a fellow researcher that have made my own tracking videos, I noted the following:
    1) The system does not seem to deal with rotation well (lost track of 3-fingered window when it was rotated). Generally, the system seems to be scale and translation robust, but not rotation.
    2) It is a single hypothesis tracker, which means temporal occlusions and multiple similar targets will cause the system to diverge.
    3) No failure cases were shown, which usually indicates a lack of testing. Everything has weaknesses (including our own vision system), where are the failure modes for the system?

    Otherwise, I like the fact that it works practically and source code is available. However, claiming that this is better than the kinect is ludicrous as it provides no 3D information (initial object scale not known with a single camera!) and can't tell the difference between the photo of a face and an actual 3D face (whereas the kinect at least has a shot at this).

  43. open source == good ... by Anonymous Coward · · Score: 0

    Right on!

  44. LOL "U GOT 'P L A Y E D'" (U played yourself) by Anonymous Coward · · Score: 0

    http://slashdot.org/comments.pl?sid=2082940&cid=35823386 , lmao. Serves you right for trolling others first for no reason Skidborg.

  45. LOL "U GOT 'P L A Y E D'" (U played yourself) by Anonymous Coward · · Score: 0

    Being the troll you are, I can see how you ask how humans do anything, first of all.

    It's clearly, beyond you, & even on forums, understanding that there is another human being on the other end, first. No, you just TROLL OTHERS, as you did myself, here AND ADMIT TO IT ALSO:

    http://slashdot.org/comments.pl?sid=2082940&cid=35823386

    Says it all, see subject-line, or for the uninitiated? Because for that?? Well...

    LOL, "U GOT 'P L A Y E D'" (U played yourself)

    APK

    P.S.=> See - Skidborg, if you wouldn't troll others here 1st for no good reason at all? You might not have gotten "PLaYeD" as badly, or rather, not at all, instead, so... Try to "drink that in & digest it" troll... apk

  46. LOL "U GOT 'P L A Y E D'" (U played yourself) by Anonymous Coward · · Score: 0

    LOL "U GOT 'P L A Y E D'" (U played yourself) -> http://slashdot.org/comments.pl?sid=2082940&cid=35823386 where skidborg admits to trolling others 1st & getting his jollies from it, like a sick troll does. Grow up, get help freak - You need it, because LOL "U GOT 'P L A Y E D'" (U played yourself).

    1. Re:LOL "U GOT 'P L A Y E D'" (U played yourself) by HeckRuler · · Score: 1

      wtf is this guy going on about? Whatever. The piano thing is from Buckminster Fuller, btw.
      My point is that emulating lifeforms is a naive way of making something. I believe most lifeforms have multiple eyes for redundancy. You know that whole thing with the two lungs, two kidney, two balls. There are a lot of lessons to learn from studying biology, but until our drones need to worry about bears trying to eat it, one camera will probably do. Especially when it's just looking at an office hallway.

  47. Re:It's not Goatse, it's a genuine ad for Audi car by fractoid · · Score: 1

    The goatse that has goatse in the link is not the true goatse.

    /zen

    --
    Rampant carbon sequestration destroyed the Dinosaurs' tropical paradise. I'm here to help repair the damage.
  48. Re:It's not Goatse, it's a genuine ad for Audi car by rubycodez · · Score: 1

    the goatse is a lie

  49. Interesting by iinlane · · Score: 1

    I've been watching progress of Yann LeCun on this topic for years now. Cool to finally see an application.

    http://www.cs.nyu.edu/~yann/research/objreco/index.html

  50. advert by Anonymous Coward · · Score: 0

    This article is written like an advertisement. Please help rewrite this article from a neutral point of view. For blatant advertising that would require a fundamental rewrite to become encyclopedic, use {{db-spam}} to mark for speedy deletion.

  51. This is a great contribution to humanity. by Anonymous Coward · · Score: 0

    I hope he can mod this to track eyeball movements and be able to use our eyes as a pointing device like a mouse. This would probably do away with our long dependency on the mouse and probably help a lot of disabled people.

    Also if it can track our hand and finger movement we can maybe use it to manipulate screens from a distance. Imagine clicking an icon in your pc screen from a distance. It can be something like the operators in the movie Avatar where they do gestures to manipulate a 3D projection/hologram.

  52. "This is how AI should work"? by Anonymous Coward · · Score: 0

    What's with submitters adding their own little opinions before anyone gets a chance to read the articles?

    Who are YOU?

  53. Re:It's not Goatse, it's a genuine ad for Audi car by webmistressrachel · · Score: 1

    This should be modded up... so far up... right up,.. not just because it's a great comment and this is the third or fourth time i've been back to read it, but because it gave me an excuse to say the above. Thank you, Poster.

    --
    This tagline was transcoded to result in at least one smirk. If you experience failure to smirk, please consult your Gen