University vending machine error reveals use of secret facial recognition | A malfunctioning vending machine at a Canadian university has inadvertently revealed that a number of them have been usin…::Snack dispenser at University of Waterloo shows facial recognition message on screen despite no prior indication
Cameras are one thing.
But if you can actually process it, that’s a meaningful cost per unit. The only reason you do that is if you’re planning to use it.
This type of analysis is cheap nowadays. You could easily fit a model to extract demographics from an image on a Jetson Nano (basically a Raspberry Pi with a GPU). Models have gotten more efficient while hardware has also gotten cheaper.
MSRP is $100. Even assuming you can cut that to $50 in bulk, $50 per unit is something that manufacturers are going to take seriously as an added cost. They’re not going to pay it without an intent to use it.
And that’s before software costs. Even leveraging open source it’s still going to take investment to tailor it to your deployment.
I doubt they would implement thing on every vending machine. They can still derive some useful analytic data from a smaller sample size
That’s using it.
The only possible reason to have the hardware is because you intend to use it.