Member Discussion

Athena Security - AI To Detect Guns

The first school in the world installs Athena Security

Athena is also working on developing software that uses radar to detect concealed weapons.

I think this is a much better idea then shot detection for the simple fact that AI can (potentially) detect a firearm before the round is down range. Any feedback on Athena's AI? 


"I think this is a much better idea then shot detection for the simple fact that AI can (potentially) detect a firearm before the round is down range."

down range implies some level of time savings.

what do you think is the time frame between display of weapon (which can then be potentially detected via analytics) and first shot?

serious question.

I don't think it is long enough to support your 'much better' analysis.

Of course, it depends on the situation whether AI will give you any more time vs shot detection.  But there is zero chance shot detection will give you any warning before a shot is fired.  That said neither system is going to stop a bad guy with a gun. 

Michael, thanks for sharing. 

There is an increasing number of startups claiming AI based gun detect. Anyvision, e.g., mentions it as one of their options. Ironyun, others who I do not immediately recall.

This is being driven by the general trend of deep learning models that categorize vast categories of objects, including guns, bikes, street signs, and many other categories (see ImageNet, 1000 sysnets / categories commonly detected).

As for this company, the co-founders previously ran a payments startup that raised $130 million before leaving the company.

The good news is that they have a general track record. The bad news is that neither has any special expertise in computer vision / machine learning that would give them an edge here.

According to a Fortune article on them, they are using a SaaS model and charging fairly significantly:

Security will be available in three tiers: Professional, Enterprise, and Unlimited Professional, with increasing feature sets. The price, paid per camera each month, ranges from $25 to $100, depending on the tier

As for the school in the article you cite, the video below explains that the co-founder's dad teaches there:

We might do an article on them if they get funding and/or start seeing some traction but, for now, they look like one of many fledgling AI startups without a clear technological advantage.

Video at 1:00 mark seems to show advanced discernment between actual and staged threats.  Perhaps the AI is picking up on the casual stances of the onlookers :)

“This is not a real gun, but if it were a video of this incident would be sent to police...”

i think that statement is more the lack of any real understanding of the technology itself by the glowingly non-skeptical reporter who voiced-over that piece with the traditional hard-hitting, concerned-sounding affect.

or something.

gnaw on this:

Imagine a school that pays for this type of fee-per-camera-per-month technology to assuage their constituencies' valid and/or invalid fears of their facility being the unlucky 'next' victim of some crazy loon on some type of crazy mission.

How long can we expect the school to pay for such a thing - when statistically nothing happens over, say, the next 5 years?

When fiscal year 2023 rolls around, and other things need addressing, what do existing customers decide when funds are scarce (which is always)?

imo - and my opinion carries no expectation of professional expertise whatsoever - anyone considering investing in this type of 'value-unproven' technology should consider the above.

How long can we expect the school to pay for such a thing - when statistically nothing happens over, say, the next 5 years?

Maybe it depends on the false alarms.  If it’s falsing every other day, nobody is going to miss it.

Paradoxicaliy, if it never has false alarms, then it may suffer the fate you indicate.

However, if every few months it gets triggered by a toy gun or other non-lethal but sufficiently gun-like object, then the exception might prove the rule, and build confidence that it is worth having.

good point about the false alarm factor... I hadn't even thought about that side of it - how that could play a role in future spending decisions.

<EDIT> I will point out, however, that if it falsing at frequent rates, they may alternatively discontinue use just based on how annoying that is....

PLUS, if they do keep it anyway, and a real situation occurred, these 'analytics crying wolf' scenarios will have a negative impact on response to a real situation.

Then there is the age old question, should you make it known to the public that such a system exists and generally what it’s capabilities are?

If you feel that the system probably won’t work in reality or will eventually get removed or shut off, then maybe you should publicize it and get whatever deterrent value you can from it :)

Maybe I'm missing something but this seems to me another attempt to profit on anxiety rather provide a real means of preventing the horrific actions of a psychotic killer. 

If a gun is drawn and then detected inside an high school, it's too late. People will be shot and people will be injured and/or be killed. 

It takes five seconds or less to fire a dozen single shot rounds. It will take law enforcement minutes or more to arrive even with weapon detection via video surveillance. 

What is the benefit of detection of a bad situation where the prevention of the undesired  outcome cannot be attained?

Some have said that it could reduce the number of fatalities due to faster response. I'm not sure on that.

Some have said that it could reduce the number of fatalities due to faster response. I'm not sure on that.

but since

It takes five seconds or less to fire a dozen single shot rounds.

every second counts, no?

Yes, every second counts but, my point is minutes don't resolve a problem occuring in seconds.

Similarly, if you record at 1 image per second, you can't resolve what occurs with slight of hand on a card table on playback.

Too little information, too late to make a difference.

Don't get me wrong, I'm all for technologies to improve safety and security. Just offering a real world concern for the efficacy of the technology presented.

Too little information, too late to make a difference.

Although I agree with your point that a (few seconds) faster response time is unlikely to prevent any casualties, I’m not sure it “doesn’t make a difference”.  

In this 2014 report on Police Response to Active Shooters, it states that half of active shooter events from the study were still ongoing when police arrived.  So in those cases every second really can count.  Btw, here is their breakdown:

Agreed, my apologies. One second and one life does matter. I didn't mean to imply otherwise. 

Just struggling with the efficiency of the proposed technology. 

Kind of like going to a gun fight with a knife. 

Of course, successful police intervention actually requires that the police intervene when they get there...

Assuming gun detection can work (not certain but let's say for argument's sake), it is still a feature of a video surveillance system and not something that can sustainably be sold by itself. In other words, to the extent this becomes a common feature, it will be either from providers selling a whole range of VMS / analytics or Athena offering such a whole range.

Paying $25 a month per camera just for gun detection is hardly worth it for anyone and many will include it for 'free' with overall analytic / VMS systems (as these things become viable).

Apply this to Cities as the customer: Drawn weapons are illegally used for more than just shooting people--that's the worst case, but lots of incidents can be prevented that lead to death ... like armed robbery, car-jacking, drug deals, other illicit use of force ... smart cities must have this type of video detection. It gives a real-time picture of the suspect to dispatch. Anyone disagree?

I think that has merit. It can be argued that in areas where such a system would be installed, there aren't any reasons to brandish a firearm unless you are committing an offense, or defending against one. Either requires police intervention.

The would be challenges however. It would likely be very difficult to accurately detect at a distance unless you had China-like camera density. Inside buildings the distances involved are much closer - low ceiling heights, narrow corridors, etc. Outdoors would be a much different animal requiring significant PPF for accurate detection.

I live in a state where 'open carry' is permitted by law. (I don't, but certainly spot people weekly that do.) There is a difference between 'brandish' and 'carry', but the difference is thin.

Remove jacket to sit.....brandish or carry?

Most certainly not brandishing.

bran•dish

wave or flourish (something, especially a weapon) as a threat or in anger or excitement.

That said I think this analytic isn't intended to alert on holstered weapons.

The police can/will arrest you for brandishing a weapon just by making it known at the wrong time; pulling back your shirt to expose a handgun tucked in your waistband is brandishing.

exactly. Judgement is a sticky thing in situations like this.

but lots of incidents can be prevented that lead to death ... like armed robbery, car-jacking, drug deals, other illicit use of force ... smart cities must have this type of video detection. It gives a real-time picture of the suspect to dispatch. Anyone disagree?

i have long been a critic of this type of gun-detection visual analytic 'solution'.

while it certainly has it's place (think convenience stores, banks, etc as an example - where guns are displayed in order to gain compliance - and get the loot) - the majority of gun crimes are accomplished within milliseconds of that same gun being pulled from wherever the perp has it hidden before they come out blasting. Anyone disagree?

I agree that this is simply another analytic to add to the toolbox, and definitely should not be relied upon too heavily.

Now that this system has been advertised to be in operation, a shooter would know to simply better conceal their firearm. Tuck it within a baggy sleeve, close contact with the victim, or even disguise it as something benign like a shoebox or hold it within a backpack with the muzzle poking out through a hole. Lots of ways to defeat this entirely visual approach.

I'm also skeptical about the ability of radar to passively detect a gun without a LOT of false alarms on otherwise benign objects.

Good point catch the criminal before they shoot is better then after.

"catch the criminal before they shoot"

this is the flaw in the logic of such an analytic, imo.

might it do this? sure, it could happen.

is this probable? I think not.

here is something to gnaw on when deciding whether these guys are legit - or not:

Surveillance Company Says It's Deploying 'Coronavirus-Detecting' Cameras in US
Athena Security previously sold a system that it claims can detect weapons in video feeds. Now it says it's applying a similar approach to spotting fevers.

Our Fever Detection COVID19 Screening System is now apart of our platform...

I bet their "platform" is about as effective as their proofreader. Do they have bridges for sale too?

Athena had live gun detection demonstrations at GSX and ISC West last year for live demonstrations. I thought it was impressive and haven't seen any other gun detection companies ever demo live.

As a general thought, I think such demonstrations are good marketing but I wouldn’t put a lot of weight into them because it is pretty easy as a company to control how the demo is done and ensure a very high degree of success. Now if people are allowed to randomly bring their own guns, that would be more impressive to me as it would be harder to control.

I saw them also, I tried to fool the system along with some of my colleagues and we could not fool it. It worked.

"I tried to fool the system along with some of my colleagues and we could not fool it."

how, exactly, did you and your colleagues attempt to fool it?

Interestingly, Athena has a significant drop in employees over the last few months, per LinkedIn:

In fact, everything is much worse. They fired 3/4 of the engineering staff at the beginning of the year. Fired people just don’t update their pages because when you have a job it’s easier to bargain with a new employer.

Currently, the company consists of 2 or 3 engineers and a support team that is filtering false alarms before they arrived at the customer's admin dashboard.

I have some experience with Athena. They seem like a bunch of used car salesmen to me. And I don't mean the certified, pre-owned kind of sales guys, I mean the "bad credit, no credit, doesn't matter - we finance anyone" kind of sales guys.

I see this as a hindrance to non uniformed law enforcement. possibly creating a larger life safety situation.