This Facial-Recognition Tech Designed To Sell Makeup Is Scarily Accurate

Sometimes you turn on the news and see a clip of an unknown perp robbing a convenience store. The problem is that police can't tell who the thief is because the video is so bad. But that might not be the case soon enough, thanks to French company Vesalis and its bang-on facial-recognition software.

Originally designed for department store kiosks, the technology is so good that the French government wants to use it for security reasons.

In its original form, the software could instantly pick up a face from a low-quality security camera feed and alert a salesperson via an iPad that a shopper had entered the building. The salesperson could then look back at the customer's picture, her name and her past purchase history, and then suggest a lipstick she might like, for example.

Because it does such a good job of pulling faces from grainy, low-res footage, the French government wants to use it to keep tabs on its catalogue of known "people of concern", which is similar to the US Department of Homeland Security's terrorist watch list. In fact, it test-drove the tech last October at a soccer game. It checked 20,000 people every 20 minutes against a database of 500 "problem individuals". It had a nearly perfect performance, with a 98 per cent accuracy rate. Although, in similar tests, the company's vice president of US sales says they've seen just a 61 per cent success rate. But all that aside, it brings up some really amazing (and perhaps scary) possibilities when it comes to catching crooks. [IEEE Spectrum]


Comments

    You wouldn't want to be without ID and be in the 2 - 39%.

    Hmm you might want to reread the source article:

    "The system checked 20 000 people every 20 minutes against a database of 500 problem individuals and had an accuracy rate of 98 percent; competitive technology, said Michael Vannier, Vice President of U.S. Sales, has had an accuracy rate of 61 percent in similar tests."

      Hello,

      This part must have been misunderstood. Let me just clarify.

      During this interview, I was referring to other solutions that were introduced in the US a few years ago and used in pilot projects at places such as the Tampa Police Department or the Boston Logan Airport. Those Facial Recognition solutions had an accuracy rate that was at around 61% .

      Our advanced software offers much better results in terms of accuracy and sets our Company apart from a lot of our competitors. We very often obtain accuracy rates that are above 90%.

        check 20000 people and see if 500 problem individual in them? sounds fishy, the system can always tells that none is in these 20000 people and still got pretty high accuracy (yeah, 98% is highly correlate to 500/20000), right? so need to specify how the test is going on?

Join the discussion!

Trending Stories Right Now