CaraComp
CaraComp
Forensic-Grade AI Face Recognition for:
Get Started7-day refund guarantee**
facial-recognitionBy Cara Candelario

Facial Recognition Technology: Hong Kong's 60,000 Cameras

facial recognition technology on a Hong Kong street, with a pedestrian under a city surveillance camera at dusk
A pedestrian passes a city camera as facial recognition technology comes to Hong Kong's street network. Illustration: CaraComp

Quick answer

Is facial recognition accurate on street camera footage?

Facial recognition is less accurate on street footage than on clean photos. It works best with clear, front-facing, well-lit faces, while street cameras deal with odd angles, shadows, crowds and motion. A result from that footage is a lead for a person to check alongside other evidence, never proof.

Picture a normal Tuesday. You walk to the shops, pick up milk, and cross at the lights. Somewhere above you, a camera may be running facial recognition on your face, and nobody asked whether you were okay with that. That is the future Hong Kong just stepped toward.

Facial recognition in Hong Kong's city camera network means a face check can now happen on an ordinary street, so the real questions are whether people are told and what happens when the system gets someone wrong.

TL;DR

Facial recognition is moving from your phone into city cameras, and the hard part isn't spotting a face. It's telling people when it's on and fixing the mistakes.

What is facial recognition on an ordinary street? What Hong Kong just switched on

Let's start with your worry, because it's a fair one. Am I being scanned? Could it pick the wrong person? Could that person be me? Short answer: facial recognition technology is real, it is spreading, and the mistakes are the part nobody advertises.

Hong Kong police run a camera system called SmartView. According to the reporting gathered for this piece, they have installed roughly 5,000 cameras since it launched in 2024. Police credit the system with helping solve nearly 2,200 cases and leading to more than 2,800 arrests. Those are big numbers, and they explain why officials want more. The plan, as reported by France24, is to grow the network to about 60,000 cameras by 2028.

5,000 → 60,000
Hong Kong's planned jump in SmartView cameras, from about 5,000 now to about 60,000 by 2028
Source: reporting summarized from France24 and Biometric Update

Is AI face matching running on everyone, all the time?

Not yet, according to police. They say the face-matching step will be switched on for specific searches, not continuously across every camera, and they blame limits on computing power. Here's the catch. A limit based on computer capacity is not a promise. Computers get cheaper and faster, and networks get bigger. A rule that exists because of hardware tends to fade when the hardware improves.

Hong Kong facial recognition: surveillance and privacy beyond Hong Kong

You might be reading this from a very different city. Fair. But Hong Kong facial recognition is a preview of a pattern: cities already own the cameras, so adding a face check is mostly a software decision. No new poles. No ribbon cutting. Just a quiet upgrade to something already watching the street. This article is part of a series, start with Fake Age Verification Steam Users Face Credit Card Scams.

That's the shift. Until now, most of us met a face check by choice. You held up your phone to unlock it. You stepped up to a passport gate. In both cases you knew it was happening. A facial recognition system built into street cameras flips that. The check comes to you, and you may never see it.

The AI functions "will naturally be applied to people, such as tracking a criminal suspect."

Chris Tang, Hong Kong's security chief, speaking to lawmakers

Notice the word "naturally." Tang, the security chief, also said the tools will watch crowds and read license plates. Reasonable, say supporters. And some Hong Kong residents shrug. One said there are already plenty of these AI-powered cameras in mainland cities and "people have gotten used to them." That's a real point of view, and it deserves respect. But getting used to something is not the same as agreeing to it.

How does a Hong Kong face check compare with a passport digital image?

This is the part that should make you pause. Facial recognition does best with clear, front-facing, well-lit pictures. Think passport photos. Streets are the opposite. Here's the contrast side by side.

SituationWhat facial recognition technology faces
Passport-style photoFace straight on, even lighting, nothing in the way, so a match is far more dependable
City camera footageOdd angles, shadows, crowds, motion, so a face check is much more likely to miss or mix people up
Phone unlockYou chose it, you know it's happening, and you can turn it off
Street camerasOften no notice at all, so the person being checked may never know

One set of figures in the research puts the real-world picture in numbers: accuracy of about 82.6% and reliability of about 74.3% on surveillance footage. I can't point you to the exact study behind those, so treat them as a rough guide. But the direction matches what experts at CSIS have analyzed: results vary a lot between vendors and between lab conditions and the real world.


Trusted by Investigators Worldwide
Run Forensic-Grade Comparisons in Seconds
Detailed facial comparison reports. Results in seconds.
Get Started
7-day refund guarantee**

What happens to a person's identity when facial recognition gets it wrong?

Short answer: a real person can have a very bad week. A Washington Post investigation in January 2025 found at least eight wrongful arrests tied to facial recognition. Seven of the people were Black. In several cases, officers had described the result as a "100% match." That phrase should worry you. Confidence from a person standing next to a screen does not make the screen right. Previously in this series: Facial Recognition Cctv Banned Yet Police Ran 1m Searches.

Now do the math Hong Kong's numbers invite. If a system is wrong even a small fraction of the time, and it is running facial recognition across 60,000 cameras, the number of false flags grows fast. Each false flag is a human being who now has to explain themselves. And who has to prove the mistake? Not the machine. The person.

Police say facial recognition is a lead, not proof. That's a good rule. But a rule only works if somebody checks the lead. Look at the officers in those eight cases. They had the same rule, and they still trusted the result.

Why Hong Kong facial recognition is a warning for any city

  • Scale outruns safeguards - going from 5,000 to 60,000 cameras is faster than most cities can build checks like bias reviews (tests for whether a system works worse on some groups) and human sign-off.
  • Notice is missing - the research found little public notification or consent in Hong Kong's rollout, a basic privacy gap, so most people can't tell when a face check is running.
  • Limits can fade - "only for specific searches" is a computing limit today, and it may not be one tomorrow.

How can anyone tell if a facial recognition match is trustworthy?

You can't tell by looking at a result on a screen. A trustworthy match comes with three things: a match score (a number for how alike two faces are), a note on how clear the picture was, and a real person who looked at both faces before anything happened. If a match arrives as just a name with no score and no human check, be skeptical.

Here's the one thing worth doing. When you read about any face-based identification, in the news or in your own life, ask: what was the match score, and who checked it? If you've ever wondered whether a photo or profile is really who it claims to be, that is the exact question this kind of facial recognition technology exists to answer. The good versions answer it with evidence a person can review. The bad ones answer it with confidence. That gap is where the wrongful arrests live. Tools built for careful investigators, such as the ones we build at CaraComp, are designed around that human-in-the-loop idea (a person always makes the final call), not around a machine's gut feeling.

As for what to expect in your own town: this kind of face check arrives quietly, usually through contracts and budget votes. Local meeting agendas are where you'll see it first. The IEEE Public Safety group has written about transparency and consent in police use, and its framing is worth a skim before your next council meeting.

Key Takeaway

The question with facial recognition on city cameras isn't whether it can name a person. It's whether people know a face check is happening and who clears them when it's wrong. Up next: Facial Recognition Hong Kong Cameras Jump To 60 000 By 2028.

Recap: Hong Kong is putting face checks into its everyday street cameras. What matters most is being told when it's on and having a fair way to fix mistakes.

So here's the thing I keep chewing on. In Hong Kong, a camera that once just recorded the street can now be asked who is on it. Walk past it every day for a year. Would you ever find out the day it started asking about you?

Facial recognition: Frequently Asked Questions

What is Hong Kong facial recognition on city cameras?

It is a system called SmartView that police are expanding. Cameras already record public streets, and the new part lets facial recognition software compare facial features in that footage against other facial images. Police say it will be switched on for a face search in specific cases, not constantly. Roughly 5,000 cameras are installed, and the plan reported by France24 is about 60,000 by 2028. Hong Kong's security chief, Chris Tang, has said the AI functions will be applied to tracking criminal suspects, but the public gets little advance notice about when it runs.

Is facial recognition accurate on street camera footage?

It is less accurate than on clean photos. Facial recognition technology works best with clear, front-facing images of faces and good lighting, which street cameras rarely provide. Facial verification, a biometric check such as unlocking your phone, compares one face to one stored photo, which is an easier job than searching crowd footage. Figures in the research put real-world accuracy around 82.6% and reliability around 74.3%, though the exact study behind them is not named, so treat them as rough. A match from street footage is a lead for a person to check, never proof by itself.

Can a facial recognition database match lead to a wrongful arrest?

Yes. A Washington Post investigation in January 2025 found at least eight wrongful arrests linked to facial recognition, and seven of those people were Black. Officers in some cases had called the result a "100% match." Facial identification can be wrong, and people can trust it too much, which puts the wrong identity on an innocent person. That is why a human review, a visible match score, and a way to challenge the result all matter before any arrest, and why they protect the rights of the person flagged.

Should people be told when a face check is happening in public?

Many experts say yes. IEEE Public Safety has written about transparency and consent in law enforcement use, and the research found that Hong Kong's rollout has little public notification. Facial surveillance in public spaces raises privacy and security questions, and it touches basic rights. Telling people when and where cameras run facial recognition, what data is kept, and why, lets residents ask questions and challenge errors. Without notice, a wrongly flagged person may never learn why they were stopped or how to clear their name.

Ready for forensic-grade facial comparison?

Full forensic reports with detailed similarity scoring. Results in seconds.

Run My First Search