The above video for the aspiring GunDetect system doesn’t show the system. Make the jump for the image. Meanwhile, its creators want you to know this — they’re not playing politics. “Our goal with GunDetect is to find common ground and avoid the debate about gun laws, since we are technologists with a technical solution that can provide an unobtrusive layer of safety. We’re particularly interested in making sure children are safe, which is a goal we share with the NRA’s Eddie Eagle program and with the numerous gun-safety advocacy groups. If our GunDetect system gives an early warning and helps save a child’s life, then all the work we’ve put into creating this product will be rewarded as the most important contribution we can make in our careers.” Here’s their presser on how it works . . .

GunDetect (courtesy gundetect.com)

GunDetect’s optical sensor needs to have a clear view of the area to be protected, and it can be set on a shelf or hung on the wall, requiring only a standard electrical wall outlet and a WiFi connection. Once purchased, consumers can register their GunDetect product online and assign a phone number to the device, which will send an alert via text message when a gun is detected in its field of view.

“We’ll also offer lower-cost versions that use our cloud-computing software on the Internet, but our top-of-the-line GunDetect integrates a powerful vision-processing system that does all of the required computation locally and then wirelessly transmits just the results – the presence or absence of a gun – via the Internet to generate a text message,” said Rajeev Kumar, VP of Marketing for GunDetect.  “This means that you don’t have to be concerned about video leaving your home – we built the system to be secure and private.”

GunDetect uses specialized vision-processing software that enhances a rapidly emerging technology called “Deep Learning,” which is the same technology that is being developed for driverless cars. The basic algorithm has been around for years (sometimes described as neural-net processing), but due to its staggering computational complexity, only recently has the high-speed hardware needed to implement Deep Learning become affordable.

Deep Learning software is “trained” by software analysis of many thousands of images that are labeled as to whether or not they contain the object that is being recognized (a gun). Once trained, the software can interpret new images to decide whether they contain the object of interest, but getting high accuracy requires some complicated image manipulation along with other vision-processing algorithms to help the deep-learning software avoid false results. The GunDetect software incorporates techniques from recent university research, in addition to some proprietary technology.

Too little too late? Something that can be used against gun owners? Or gang bangers? Or an apolitical gun safety idea that begs for adoption? Or all of the above? Check it out:

Test GunDetect! Even before GunDetect hardware enters final production, you can try out the latest software, which is running on the GunDetect cloud server. Sign up at www.gundetect.com and submit your own image to test GunDetect’s ability to identify a gun.

Pricing and Availability:

Versions of GunDetect range in price from $399 to $800, with deliveries for early Kickstarter backers scheduled for Q1-2016 and retail production planned for late 2016.

75 COMMENTS

        • Only if the box contains long stem roses carried by a 6’6″ Goliath wearing black leather and black sunglasses.

    • Good question, but I wouldn’t trust my life or anyone else’s to an $800 neural network. Neural nets are good at pattern recognition, like being able to tell whether a shape is an apple or a fork. To “train” them, you feed in a bunch of images of apples and forks until the network can reliably identify one or the other. Then, BAM!, you have a neural network that can distinguish apples and forks more often than not. Until part of the apple is obscured, and then it’s neither an apple nor a fork. Or until it sees a peach for the first time and thinks it’s an apple.

      I’m oversimplifying, but being able to get an $800 piece of hardware to reliably spot a gun with today’s hardware and [more importantly] software, in real time (or not), is purely science fiction. It would still be science fiction at 1,000,000 times the cost. If it were possible, wouldn’t the military have already built one to spot IEDs or bomb vests or I don’t know, GUNS??? Oh, right, science fiction. If anyone could actually build such a system, they’d skip the consumer nonsense and sell it to the government for ten hundred bazillionteen dollars.

      TL;DR: What a joke.

      • Compsci Guy, thanks for the technical post, and we’d welcome a discussion of the technology challenges we’re overcoming with GunDetect. I’m the CEO of the company developing GunDetect, and I can provide Robert and his readers with some more information about the vision-processing algorithms we’re using. As you imply, there is a reason why nobody has built a vision-based gun-detection system before — especially within the price points required by consumers. The early neural-net algorithms have been superseded by some new approaches that dramatically increase the accuracy, though you’re certainly correct about the limitations. With our current software and clean images, we’re getting around 90% accuracy, and we’re now improving it with some pyramid approaches to help the classifier.

        We’re still finalizing the user interface for those GunDetect users who want to help improve the learning system, but I think you’ll like this approach. We’ll allows owners to optionally supply the GunDetect with a text reply that trains GunDetect to learn from any false positives or false negatives. This is already happening with our web-based demo at http://www.gundetect.com. After you submit an image, you can click a box to say whether the software correctly identified the image. All of this feeds into our training algorithms. Basically, GunDetect will keep getting better over time as the owners use it more, and the improvements will automatically roll out to all the units with an over-the-air update. I encourage you to give our demo a try and then compare it to any other vision software that exists.

        BTW, we’re requesting that users register on our site, but we’re only collecting emails so we can contact people about the Kickstarter — which should launch next week. We also need the physical address because we’re going to face some export-control issues. As you might expect, the login requirement has the added benefit of helping to throttle access to our cloud-based servers, since these are expensive GPU machines designed for high-performance computing.

        Feel free to join the discussion in the forums on gundetect.com or share your thoughts to @gundetect on Twitter. I’ll also stay active on Robert Farago’s site in these forums.

        Thanks,
        Scott Gardner
        CEO, GunDetect

        • Scott, can you describe a few scenarios where gun detect would be useful? I don’t dispute that you can detect guns, but I find it hard to believe that there are enough circumstances where it will be useful. The gun must be in line of sight, at a workable angle to the lens, all without any object occluding the gun. So if the child holding the gun has its back to the lens, it won’t work. Please explain.

          • Gary, you’ve correctly identified all of the constraints for a vision-based system. The GunDetect needs a clear view of the gun in order to detect it, so any obstructions would be a problem. We believe the most likely scenario is for gun owners to position GunDetect to watch a drawer or cabinet where they have a gun stored — preferably from a relatively close distance at an angle that keeps the gun in its field of view. We also believe that users would like to be notified if someone takes a gun out of the vault, especially if the person is only supposed to open the gun vault in an emergency.

            To those owners asking about open carry, we expect that most people keep their weapon holstered, which we’re training GunDetect to ignore. An alert gets triggered when the system sees a gun in someone’s hand, which is something that we’re assuming the GunDetect owner wants to know about — in a home, church, business, whatever. They can always ignore the text alert if everything is fine.

            Scott

  1. Uh, how about training your children? You know, actually spending the time to familiarize them on safe firearm practices? And not leaving them home, alone? Radical, no?

    • If I ever found out a place had one of these, I would go set it off three times a day.

  2. As someone who has extensively worked with top of the line industrial vision systems, the technology they are describing does not exist. If it did, I’d be out of a job.

    • Yup. Not commenting otherwise, I certainly can see where it could be useful, but this software is vaporware.

      • Vaporware that will leave the promoters well off financially after the inevitable bankruptcy. Do they actually teach this kind of racket as a business model in some B-schools nowadays? Seems like it; the pattern is so familiar.

  3. What if it is detecting and vilifying legal carry people? Who would regulate proper use so as not to break into Rights infringement? And to throw the “children” word into the mix as a sales method? Why did they have to go there? My gut says this will be used improperly. It will not satisfy gun haters, and will never be in the right place at the right time to stop the very small portion of the population who make poor choices. Haters will fight no matter what we do to display responsibility. ANSWER: No, I’m against the product.

      • Why?
        I see this thing crying Wolf on a lot of innocent people who will be stopped and frisked for a trumped up probable cause.

  4. I smell a scam. Like most Kickstarters, its just a way to fleece some morons for cash. Thanks to Kickstarer, Televangelists will be a thing of the past soon.

    • If they are attempting to fleece morons for cash, they picked a pretty good group for their target.

    • I see no problem with taking anti-gun loosers’ money… You’re really just taking back the money you paid in taxes for their welfare checks. Where’s the problem? You get back the money they stole. Kinda like being a slumlord…

      I considered a website where PETA/Lefty losers could crowdfund my $1000/mo goal to NOT go hunting. Why not? If they really love Thumper and Bambi so much, is $1/mo from 1000 of them too much to ask? Oh, I guess even their own opinions aren’t worth $1 to themselves, interesting…

      • Hmmm.. good idea, considering it costs a lot more to not hunt (200-300 practice/target rounds costs more than hunting all weekend and taking 6 shots) I’m game for that, so to speak. I hereby promise to not hunt with any ammunition paid for by PETA donations to me.

    • Maybe we should warn MDA that it is scam to take their money and produce no useful product. /Sarc

  5. Can it tell the difference between a J-frame and wallet in someone’s front pocket? I doubt it.

    I smell a scam.

  6. Someone should submit that picture of Shannon Watts next to the TTAG rep… see if it detects the gun, then she can just hang one of these around her neck.

    • Uh oh. She might demand I register Mini-Me as a lethal weapon if it gets too close to her. But it cannot help itself! Those crossed-eyes, that blank stare, the curled lips and elongated neck accentuated with arms crossed and the body’s weight shifted on one foot.

      • You left out her flat face and man hands…

        ‘jus sayi’n…

        (I know, I know, she makes you happy…)

      • Dude! When did you get your hands on an FN mini-me? I have been wanting one of those for some time. Can you hook me up with your dealer of such thing? I……wait you are talking about something else; aren’t you?

  7. Well it seemed too good to be true, but after seeing the commercial, I’m convinced!!

    *sarc/off*

  8. If you have an untucked shirt with a firearm not printing this gizmo won’t detect it unless it is measuring some other kind of radiation.

  9. Who hasn’t heard about all the money wasted on Shot-Spotter systems. I’ve heard of at least one locally (Brentwood., NY) that was taken out of service / abandoned because of false positives from fireworks and other common noises.

  10. “assign a phone number to the device, which will send an alert via text message when a gun is detected”
    To you and who else?…

  11. Can it tell the difference between a wallet and a gun? If so, please supply numerous GunDetects to the NYPD.

  12. ” If our GunDetect system gives an early warning and helps save a child’s life, then all the work we’ve put into creating this product will be rewarded as the most important contribution we can make in our careers.””

    That’s a huge IF. And IF this product pitch is any indication of how you envision it working then our definitions of “early warning” are very different. From what I can tell, your only reward is going to be a keep-sake video of your child blowing their face off.

  13. Some things shouldn’t be automated. There’s no substitute for common-sense. Those who don’t have any would rather buy snake oil than think for 4 consecutive seconds… Pretty much the definition of the entire leftist platform…

    No brains, refuse to get any, make others suffer because reasons.

    • Amended: Seems I was supposed to register prior to uploading. I’m too lazy to register to test this …..

      • Ilinois_Minion, I’d encourage you to work up the energy to register, since we mainly just need a validated email address to enable access to our cloud servers. I tried your image of the holstered gun, and GunDetect correctly said there wasn’t a “gun” in the image. We’re training the system to ignore holstered weapons and only trigger on weapons in someone’s hand.

  14. Assuming this works, the gun has to be in its line of sight. The child has to open the gun box in the same room without his body or anything else in the way. If the gun is detected, the parent has to be close enough to grab the gun in time. It would be cheaper and work better to use a stronger, anchored lock box.

  15. Another completely useless gadget that won’t sell well enough to keep the company afloat.

    And so it goes.

  16. If it actually goes on sale, using saving kids as a ploy, they are inviting hugh lawsuits when a kid dies or is hurt due to a gunshot.

  17. This is going to be a tool to produce probable cause for stop and frisk of the citizenry by the Thought Police. Another day in Oceania.

  18. Why would I buy a product that went off every time I entered the room?

    I can see the advertising campaign now: “It’s ten o’clock. Do you know where your guns are?”

  19. “submit your own image to test GunDetect’s ability to identify a gun.”

    Sure buddy, I’ll submit a pic of my burner! That would work better and faster than a nation wide registration scheme.
    Oh boy! There is fraud or the co. goes chapter 11 and now the gubmint has these important records.

    Tinfoil hat dawned? dunno.

  20. Their “deep learning” expert, with a PhD from Arkansas (the Stanford of the Ozarks?) doesn’t have any published papers that I can find. I highly doubt they have any tech more advanced than the simple shape recognition that is commonplace on image search websites.

    I think for preventing accidental shootings, I’ll trust my safes over a text message. And for detecting an armed intruder, I’d go with motion sensors that alert me of any intruder, rather than a scanner that will let an axe murderer walk by without a peep.

  21. Put it on a food cart on New York streets. Have it call the police every time a cop walks by and report someone with a gun. See how long it takes them to figure it out.

    • Ray, that actually is pretty funny and something we haven’t tried yet. 🙂

      We did have to train GunDetect to stop recognizing finger guns, since that seemed like a good way for kids to annoy their parents by generating false alarms. The system also does a pretty good job of not detecting water guns, but we’re still trying to decide whether to allow realistic toy guns to send an alarm — such as the Airsoft gun you guys correctly identified from the video.

      Feel free to test pictures of your rifle OR your gun to our web demo…

      Thanks,
      Scott
      CEO, GunDetect

  22. This has a strong odor of snake oil. A cheap webcam and some “secret sauce” image processing software sounds about as credible as that green coffee bean stuff. I don’t believe this does what it claims, nor do I see any real use for it even if it did work.

  23. So it can recognise guns in the open (presumaby a convolutional neural net), probably not a ccwer i imagine that would be harder

  24. And how much of this information do they communicate to the BATFE or other gov’t agency’s to create a registration file?

    • CCDWGuy,

      I’m the CEO of GunDetect, and I wanted to address this question of privacy, since we expected it to come up. We put a lot of effort into our design to make sure that everyone’s information is private. The GunDetect belongs to the owner, and they are in complete control of their information. We don’t share information with anybody, and we have zero contact with government agencies. In fact, we specifically do not save images, so that it is impossible for law enforcement to get a visual record of what GunDetect sees, just in case an owner did shoot an intruder in front of GunDetect. Other webcams, such as Dropcam and others, are recording information in the cloud. While we’re introducing some less-capable and lower-priced GunDetect units that require our cloud software, our top-of-the-line GunDetect does all its processing locally and does not send any images to the Internet — making it inherently hack-proof, in addition to protecting your privacy.

      We welcome the feedback and any ideas you guys might suggest for how we could prove conclusively that gun-owners have nothing to fear from this technology.

      Thanks,
      Scott Gardner
      CEO, GunDetect

  25. I need to figure out a kickstarter/crowdfunding scheme…er scam-no wait life-saving technology. If it can save save just one kid with an airsoft gun…

  26. This is a stupid waste of perfectly good technology… Any gun that is concealed or disguised in any way will be completely “invisible”, and the only thing it could possibly do is provide a false sense of security. Most robbers will conceal their weapons until they want to get the attention of their victims, and I think that any situation other than a business looking to protect itself would justify a full-fledged metal detector.

  27. Their test drive does not even work right. I can see this product working the same way.

  28. Unsurprisingly, there is ‘mandatory registration’ before the demo works. Oh hang on, someone’s at the door…

  29. But will it recognize a Borchardt? Also, in what way is a gun webcam useful inside your home? Talk about too little, too late.

  30. They keep talking about saving just one child’s life. However, abortion is still legal. Until they start addressing THAT issue, they will only get a giant STFU from me. Oh yeah, and POOLS! Pools are still largely unregulated. More children die in POOL accidents than firearms accidents by a large margin every year. When are they going to deal with THAT threat?

  31. I’m not totally against this. Don’t see the problem if somebody with a CCW is caring reliably and concealed hears an alarm that has a certain tone to it that they would immediately know that there’s possibly a threat nearby and they should be ready.
    People on here are saying that it can’t be done, and it’s science fiction. They have retna scanners that can read someone’s eye around 80 yards away, and your saying that they can’t train a system to figure out a gun shape? Come on, they made a necklace that can read and find videos for you real time from a book or newspaper.
    I for one honestly don’t see a problem with this technology, if I’m being a responsible gun owner I have nothing to fear.

  32. Why on earth are they mishandling that poor FiveSeven, WTH?!? Surely it has done nothing to deserve to be gripped in such a poor manner.

  33. The question in my mind is, assuming this device is trained to detect firearms based on interpretation of images from exposed or partially exposed firearms, where would it be used? If it is used in a legal open carry environment, what’s the point other than harassment? And otherwise, aren’t there enough busybodies out there already reacting to someone walking around with a gun? Or the perceived sound of a gun for that matter?

  34. I don’t see a problem with this – especially for use in a company lobby or school entrance. Even if someone in the lobby does unfortunately gets shot – it might well give any number of other people time to escape or return fire if armed.

  35. Servers are overloaded, registration doesn’t seem to work, we’re off to a great start.

    Can’t even get the test image to work and I’m assuming my own image doesn’t upload cuz of the message about servers being overloaded.

Comments are closed.