I reckon the ShotSpotter gunshot detection system is a huge waste of money. TTAG’s resident war hero Jon Wayne Taylor agrees — having seen the system’s lack of value in Afghanistan. Jeremy disagrees. No matter how you slice it, the audio-based system could soon be obsolete.

From securitysales.com:

Researchers at the University of Granada (UGR) have designed a computer algorithm that can identify guns on video in real time.

The researchers trained the artificial intelligence on low-quality videos from YouTube and movies from the 90s, such as Pulp Fiction, Mission Impossible and James Bond.

The algorithm showed an effectiveness of over 96.5%,  detecting guns with high precision by analyzing 5 frames per second, or in real time.

When a handgun appears in the image, the system sends an alert in the form of a red box on the screen where the gun is located.

UGR professor Francisco Herrera says the artificial intelligence can be easily combined with an alarm system and implemented inexpensively using video cameras and a computer with moderately high capacities.

He also says the system can be implemented in any area where video cameras can be placed, indoors or outdoors, and does not require direct human supervision.

Well he would say that, wouldn’t he? And any AI-based firearms detection device would have to be based on video surveillance (obvs.). So unless America is heading down the cameras-everywhere path trod by the United Kingdom — the most surveilled country on planet earth — video AI gun detection’s utility would be strictly limited.

That’s without taking into account the fact that most armed robberies/attacks are over in minutes, if not seconds. Even if the AI gun detection video system sends out a red alert to the po-po, the chances of the cavalry arriving in time to do anything about an armed aggression, save mopping up, are vanishingly small.

Besides, 911. But never underestimate the power of security theater! And feel-good politics (i.e., doing something). Not to mention ShotSpotter’s highly paid lobbyists. This may be one to watch. So to speak.

23 COMMENTS

  1. And how much do we taxpayers have to fork out for the privilege of being spied on or protected? I see another tax increase or a brand new tax entitled Public Safety Video Tax or something.

  2. Ok, so a would be mugger puts his gun inside of a paper bag and points the bag at someone. He could even cut the bottom of the bag off so the victim could see the gun, but the Orwellian eye could not. Could also use a shirt. Really anything.

  3. Heh. Skin tone cerakote; spots on side like a cell phone keypad… Will be trivial to fool such a system.

  4. Drape a towel, jacket, or put a paper bag over the gun and all that programming is for nothing.

  5. Ok, so how good is it at avoiding false positives? That’s the big question here. If you program it with a rather high margin of error for what it flags as a gun, then sure, it’ll be good at identifying guns. It’ll also be good at mistaking all kinds of other things for guns

    • False positives was my first thought. Can it distinguish between an AirSoft gun and the S&W it copies? What about the 1911-style squirt guns I saw kids playing with just down the street?

      If not, there’s a way to protest its installation: get some buddies, pool some money, and give kids good toy guns that will set off the software if they’ll promise to walk by the cameras with them often.

      For that matter, what about a good image of a gun? If the system trained on movies, it isn’t 3-D capable, so just walking by with a picture of a gun would set it off.

  6. Likely it would do better as an augment, rather than replacement, for shot spotter, assuming there is value in the latter. (I’m agnostic.)

    A question I have is, how does the ID reliability scale with camera resolution and lens type?

    And of course this emphasizes the point that … concealed means concealed.

  7. You haven’t been keeping up Obumer DOJ grant $ have deployed massive qty of gov’t video in cities across the country For the children. (Chicago leads the statist herd).

    ShotCrapper was/is just grantbait. 1/2 assed concept that would only work if just poured MORE $ into it. It’s only benefit was that it is expensive (and just need MORE $).

    This V2 – Video will be a great way to further rationalize the BS FirstNet previous mentioned. It sees a gun and any face within 10ft – felony/send on the warrant.

  8. Just to be clear, I think Shot Spotter does what it was made to do: identify a gunshot, pinpoint the location, and send out an alert. That’s it. Whether or not there’s value in that is ENTIRELY in the lap of the city that’s using it. Where I lived (read the article RF linked to), the police were on constant patrol, received Shot Spotter notifications immediately (routed directly to their vehicle computers), and responded immediately, treating Shot Spotter alarms like full-on emergencies. They caught people. It helped.

    If an agency doesn’t respond like this it’s a total waste of money. You get data, not results. The system does only one thing, and how you use it is the only thing that determines success or compete waste.

    • They caught people?
      The police do that now. The problem is, they don’t catch many during the commission of a crime; instead, they (sometimes) catch them later.
      Just what percentage of the shotspotter alerts resulted in catching someone at the scene? The right someone? I’ll bet figures on that aren’t available, for “security reasons.”
      If the police are actually cruising, just waiting for that alert, how expensive is that? Or do they do something else while waiting? When they get the alert, do they stop taking the report they are working on, or finish it? What’s the average response time?
      Without knowing those answers, anecdotal evidence remains anecdotal.

  9. Quoted: The algorithm showed an effectiveness of over 96.5%, detecting guns with high precision by analyzing 5 frames per second, or in real time.

    Video is more of the 24 to 30+ frames per second so this is not anything approaching real time as it’s analyzing at something like 1/5 or worse of “real time”.

    • I’m going to go out on a limb here and say it probably doesn’t analyze every single frame. If it’s doing 5 per second it’s probably grabbing every sixth frame or some similar rate. No need to analyze every single frame anyhow if you can grab a decent still a few times a second. The image isn’t going to change that much anyhow.

      Also, what difference does it make? The bad guy will be long gone before anyone is able to do anything about what it detects.

  10. KA-CHING! Someone has learned from Shotspotter’s marketing, and that’s all that matters. Because this is, improbably, even dumber of an idea.

  11. What’s going to happen when some Unarmed Black Teen gets shot as a result of the video system interpreting some kid’s cased cellphone as an M&P Shield or similar and calling in the cavalry?

    EDIT: And, as the article point out, we don’t need to know where people have guns. Most gun-owners are eminently law-abiding, peaceful people. What we need to know is where people are shooting one another, which is probably better served through the audio channels.

  12. ShotSpotter is obviously more valuable because audio detection can cover a greater area than focused cameras. Until we do have cameras everywhere – and yes, we are going in that direction with shows like CBS’ Hunted propagandizing for it – audio detection would make more sense.

    That AI pattern recognition can recognize a gun is not surprising – guns are relatively simple fixed shapes. Recognizing them while they are held in a fairly static position – pointed at someone – is also not that difficult.

    Also it depends on the camera angles and whether there is anything blocking the view – the same problem that afflicts public face recognition – put a hoodie on and it’s next to worthless (although they are making progress with recognizing people from just certain unique sections of the face.). As a poster says above, put the gun in a bag and the AI is worthless. ShotSpotter is not affected by that.

    The main problem with ShotSpotter is the same as cops in general – it’s only good after the fact of shooting.

    • The number of IR cameras needed to cover every dark alley and corner of a building would make even Bloomberg balk, yet alone a city.

      The only obvious application would be in a well-lit bank so that police can get an alert BEFORE a gun goes off.

  13. Is it designed to detect carried guns or just drawn guns?
    If the former, a string of alerts will be following every cop around town.

  14. All well and good. But in Chiraq 102 were shot over the July 4th “holiday”‘despite being flooded with cops and feds. Homie don’t care ’cause most are teen thugs with no boundaries or parental guidance… Yep security theatre.

  15. For every action there is a counter measure. Mount a high power IR led on the gun, put one on the brim of your baseball cap while you are at it and poof nothing gets made. Simple as that.

    Speaking of false positives … could someone play that campus carry demonstration video where they were carying dildoes not guns through this AI. Bet that might go viral it would be so funny.

Comments are closed.