China Facial Recognition: Hong Kong Cameras Now Match Faces

Police in Hong Kong found an elderly man with dementia within an hour and a half of his family reporting him missing. A camera network helped find him. And that same technology comes with a catch that has nothing to do with the cameras.

If you've ever walked past a security camera, this

If you've ever walked past a security camera, this story includes you. Hong Kong police say they'll switch on facial recognition across their SmartView camera network this month. They plan to grow it to sixty thousand cameras by twenty twenty-eight. The system works two ways. Police can search old footage for a suspect's face. Or a family can hand over a photo of a missing relative, and the system alerts police if that person appears on camera. So the real question is this. When a computer says "that's the person," who checks it?

Start with the good news, because it's real. In pilot tests, police say the system helped them arrest a robbery suspect. It also helped them find that man with dementia. For a family waiting by the phone, ninety minutes is a gift.

Now the numbers behind the algorithms. The article cites N.I.S.T., the U.S. government's testing agency for this technology. The sixteen top algorithms on its leaderboard miss a true match less than one time in a hundred. And they flag the wrong person roughly three times in a thousand. Those sound like numbers you can trust.

But those tests use clean, well-lit photos. The article notes that algorithms struggle with the blurry, grainy images police often have. A camera on a rainy street at night isn't a passport photo. The same N.I.S.T. testing found that wrong matches can run up to a hundred times higher for Black and Asian faces than for white men. That's a hundred times. In a city like Hong Kong, that gap matters.

Hong Kong built in some friction

So Hong Kong built in some friction. The system won't run on every camera at once, because continuous analysis takes heavy computing power. And an officer ranked inspector or above has to sign off before it's used. Every result also needs review by an officer trained in comparing faces. That's a person looking at the match, not just trusting the score. It's a speed bump on purpose.

One more detail. Facial data will be kept longer than the usual thirty days. Sources point to about a hundred and eighty days. That gives detectives more to work with. It also means your face sits in a system for months. For police, that's a bigger toolbox. For everyone else, it's a longer paper trail.

Is this just surveillance creeping forward? The article gives the other side fairly. Almost eighty percent of people say they're comfortable with police using Ai Voice Cloning 3 Men Charged In 129 Million Boss Scam to search watchlists. And the cameras already existed. Supporters say this speeds up detective work rather than creating something new.

The weak link isn't the algorithm. It's the person reading the result. The article says some officers treat a computer's guess as proof, even calling an unchecked match a "hundred percent match." No software update fixes that.