I don't think you understand the problem scope. Guns are hidden behind clothing and you're going to have a lot of false positives.

In this instance, it certainly would have worked.

“One man jumped onto the teller counter and pulled out a handgun, throwing down a garbage bag for the tellers to fill with money”

Outside cameras could have tracked them into a car, if it was reasonably close.

A security firm can monitor for false positive.

This would probably work for your average convenience store robbery too. In the US, we have a large market for security.

Perhaps you would be suprused by the number of people who carry guns 24/7. This would be very hard to implement.

Perhaps you'd be surprised that some countries with banks have very few guns in private possession.

In countries that aren't the USA, that number is functionally close to zero.

Private businesses, especially banks, are free to ban guns (and may be required to do so). So while this might not work for the gas station next to the hunting club, it should be fine for the bank.

I didn’t say it would be easy. I said there’s probably a market for it.

As with any real world problem, there are lots of obstacles. Sure, your AI will need to figure out if the gun is being pointed at someone, etc.

Most are concealed, and the rest are usually holstered and pointing down. Doesn't seem that hard to exclude almost all of those, leaving just a few for manual review.

Not only recognition, but also video quality adjustments. A bank surveillance system often has to archive months of video so police can go back to try to find out if the robbers had been there before to plan their crime.

A smart AI surveillance system might have 4k cameras, but not save the video in 4k format.

Instead, everytime a new face appears, it can use the maximum resolution to take a profile photo of the face for archival purposes and to calculate facial recognition.

If integrated into the banks software, it could match the faces to associated accounts by noting what account was looked up while the new face was at a tellers window.

If a robbery is detected through behavioral analysis or gun recognition, then the recording system would switch from 720p @ 15 fps (for example) to 4k @ 60 fps. It could also not just alert the police automatically, but could quickly pull up the associated account information, clips of previous visits discovered through the facial recognition, and send that to the police or have it ready for their viewing.

How exactly does that help? A silent alarm button was probably pushed by an employee anyway.

And whatever cameras the bank had, they were already recording.

How long did it take for the employee take to push the silent alarm? Seconds count. (I realize you have no idea and simply made that up anyway. We don't know if someone did push the alarm.)

Look at the grainy pictures.

https://fox6now.com/2018/10/16/recognize-them-hartland-polic...

"Our system zooms in on people who carry firearms. In addition, our network of cameras coordinate and will take several pictures from different angles and zoom levels. Furthermore, our AI looks for identifying features (e.g. moles) on the suspect's face, arms, etc"

How many bank robberies and convenience store robberies are there every year in the US? Why stop there? There's lots of gun violence in the US that lends itself to AI based security.

Who cares about bank and store robberies? Employees hand over the money and clock out; customers continue on with their lives. Very few humans suffer, and the ones that do suffer can afford a haircut.

A better use of AI based law enforcement might be targeted against corporate+government institutions which actually harm people at scale.

Bank and store owners, for one.

Build whatever business you want. Thousands of companies are started every year.

Who are your customers?

Obviously it's not as simple as just spotting a gun. Some people carry openly.