Facial recognition: Hong Kong Cameras Jump to 60,000 by 2028
Quick answer
How accurate is facial recognition on street surveillance cameras?
Facial recognition is less accurate on street cameras than on clean photos like passport pictures. Odd angles, harsh shadows and crowds hurt performance, and the cited figures put accuracy on real surveillance footage at about 83 percent. A match is a lead to check with a human reviewer, not proof.
Facial recognition: Hong Kong Cameras Jump to 60,000 by 2028
This episode is based on our article:
Read the full article →Facial recognition: Hong Kong Cameras Jump to 60,000 by 2028
Full Episode Transcript
Hong Kong police say they've installed about five thousand cameras with a system called SmartView. They credit it with helping solve nearly twenty-two hundred cases. Now they plan to grow that network to sixty thousand cameras by twenty twenty-eight.
If you've ever walked down a busy street, ridden a
If you've ever walked down a busy street, ridden a train, or waited at a crosswalk, this story is about you. Cameras like these don't ask who you are. They just look at your face, and software decides who it thinks you are.
Here's what's happening. Police say the cameras have led to more than twenty-eight hundred arrests since the system launched in twenty twenty-four. Officials say facial recognition will switch on for specific searches, not run constantly on every camera. They give a practical reason. There's only so much computing power to go around.
But that limit won't last as the network grows. So the real question is this. When a city gets this many eyes, who pays the price when the software gets someone wrong?
Start with how these systems perform. The article's research says facial recognition works best on clean, controlled photos. Think passport pictures. Straight-on, even light, nothing blocking the face.
Street cameras don't get that
Street cameras don't get that. They get odd angles, harsh shadows, and crowds. According to the figures in the article, accuracy on real-world surveillance footage drops to about eighty-three percent. Reliability falls lower, to roughly three-quarters.
That means a meaningful share of the time, the system is off. For a detective, that's a reason to double-check. For the rest of us, it means a camera's confident answer isn't the same as a correct one.
Now scale that up. A small error rate across five thousand cameras is one thing. Across sixty thousand cameras, scanning millions of faces, it produces a steady stream of wrong flags. Each one is a real person who just happened to walk by.
We know what that looks like in practice. A Washington Post investigation in January twenty twenty-five found at least eight wrongful arrests in the United States. Seven of the people arrested were Black. In those cases, officers had described the result as a "one hundred percent match."
Read that again
Read that again. One hundred percent. And it was wrong.
That tells us something important. Confidence on the screen, or in an officer's voice, doesn't prove anything. The match is only as good as the photo, the lighting, and the person checking it.
Hong Kong's side of the story is worth hearing too. Security chief Chris Tang told lawmakers the A.I. tools will watch crowds and read license plates. He said they will "naturally be applied to people, such as tracking a criminal suspect." Some residents say they're not bothered. They point out that mainland Chinese cities already have plenty of these cameras, and people have gotten used to them.
That's a real view, and it deserves respect. But getting used to something isn't the same as it being checked. The article's expert analysis says the safeguards haven't kept pace. Those include audits for bias, required human review, and clear public notice about when faces are being scanned.
The Bottom Line
For police departments, that's a policy gap. For you, it's a simple question. If a camera flags you by mistake, would you even know? And who would you ask to fix it?
The debate isn't really about whether Whatsapp Identity Verification 4 Ways Scammers Slip Past can pick out a face. It can, under good conditions. The question is whether a tool built for clean photos stays trustworthy when it's pointed at a whole city, and whether the safeguards grow as fast as the cameras do.
Hong Kong is adding tens of thousands of cameras that match faces. On real streets, those matches are wrong more often than on clean photos. So a human has to check every one before anyone gets blamed.
Whether you investigate cases for a living or just walk to work, the lesson is the same. A computer saying "match" is where the checking starts, not where it ends. The full story's in the description if you want the deep dive.
Ready for forensic-grade facial comparison?
Full forensic reports with detailed similarity scoring. Results in seconds.
Run My First SearchMore Episodes
Facial Recognition CCTV: Banned, Yet Police Ran 1M Searches
France ran about one million facial recognition searches through its police database in a single year. That's the same France that just extended a ban on facial recognition in public cameras. Both thi
PodcastPhoto Forensics: 4 Hidden Tests That Expose a Fake Picture
In 2015, the World Press Photo Contest disqualified twenty-two entrants for manipulating their pictures. One of them had won the overall prize. The judges were experts, and they still got fooled by th
PodcastFake age verification: Steam users face credit card scams
Someone gets a message that looks like it's from Steam. It says verify your age now, or lose access to your games. The link asks for a credit card number. It says the charge is only a temporary hold.
