Mass Surveillance by Country: The Data and the Public Opinion

Counting the cameras

There is no global register of surveillance cameras. The most-cited attempt to build one is Comparitech's ongoing study, which assembles city-level counts from government procurement records, published tenders and press reporting, then divides by population. It is the best public dataset available and it is still an estimate — the authors say so themselves.

Comparitech counts cities rather than countries, so the fairest way to read it by country is to take each nation's most heavily watched city. On that basis India is not merely ahead, it is in a different bracket — Hyderabad alone runs at more than three times Seoul, the next city down.

• India · Hyderabad 79.38
• South Korea · Seoul 24.28
• Russia · Moscow 19.63
• Singapore · city-state 18.35
• United Kingdom · London 13.4
• United States · Los Angeles 12.4

Two things stand out. The first is how close London and Los Angeles actually are — 13.4 against 12.4 per 1,000 — despite Britain's reputation as the most camera-covered country in the West. The second is China, which the study puts at 494 cameras per 1,000 people nationally. That single figure drags the world average from 5.82 up to roughly 140, which is why it is quoted separately here rather than folded into the chart.

How people feel about it

The best cross-country comparison of attitudes is a survey by Genia Kostka, Léa Steinacker and Miriam Meckel, published in Public Understanding of Science and available here. It asked roughly 6,000 people across China, Germany, the United Kingdom and the United States how acceptable they found facial recognition technology.

One thing to hold onto: this was fielded in 2019. It is the most complete four-country comparison that exists, and it is also six years old. Treat it as a baseline, not as this morning's mood.

• China 67%
• United States 50%
• United Kingdom 48%
• Germany 38%

The relationship runs backwards

Put the two charts beside each other and the expected pattern simply is not there. Germany has the fewest cameras of the four countries surveyed and the strongest objection to them. China has the most of both cameras and acceptance. Britain is among the most watched populations on earth and sits mid-table on how it feels about that.

There are at least two honest readings of this, and the data does not settle between them.

The first is habituation: people stop noticing infrastructure they see every day, and objection fades as cameras become ordinary street furniture. The second is that both the cameras and the acceptance are downstream of something else — trust in the institution doing the watching. The study's own authors lean toward the second. They found that willingness to accept the technology in public tracked with trust in government and with a general affinity for technology, rather than with how much of it was already installed.

It is not the camera. It is who is behind it.

The most consistent finding across every survey here is that people do not have one opinion about facial recognition. They have several, and which one surfaces depends entirely on who is operating it.

Pew Research Center asked 4,272 Americans in June 2019 whether they trusted various groups to use facial recognition responsibly. The same technology, three operators, a 38-point spread:

• Law enforcement 56%
• Technology companies 36%
• Advertisers 18%

The United Kingdom shows the same split from the other direction. The Ada Lovelace Institute surveyed 4,109 people in 2019. Just under half — 49% — supported police use of facial recognition where appropriate safeguards were in place. Move the same technology into other hands and the support collapses.

• Used by shops to track customers 77%
• Used in schools 67%
• Used on public transport 61%

A population that splits evenly on police use rejects commercial use more than three to one. That is not a public with a vague unease about being seen. It is a public drawing a fairly precise line about consent and purpose — and the line falls almost exactly where the operator stops being accountable to them.

Support is not the same as comfort

Read only the headline numbers and you would conclude the public is broadly relaxed about this. That conclusion does not survive contact with the follow-up questions.

Pew returned to the subject in March 2022. On the top-line question, police use of facial recognition drew 46% saying it is a good idea for society against 27% saying bad — favourable, but nowhere near a majority, and with 27% unwilling to say either way. Then the same respondents were asked what they expected to actually happen if police used it widely.

• Police could track everyone’s location at all times 69%
• Would be used more in Black and Hispanic neighbourhoods 66%
• Crime rates would stay about the same 57%

Those three findings sit inside the same survey as the 46%. A public that leans mildly favourable on police use also, by a margin of more than two to one, expects that use to enable location tracking of everyone, expects it to fall hardest on Black and Hispanic neighbourhoods, and does not expect it to reduce crime. That is not enthusiasm. It is closer to resignation.

Move away from policing entirely and the balance tips outright:

• Social media auto-identifying people in photos 57%
• Employers tracking staff attendance 50%

The most recent figure in this piece points the other way, and it belongs here for exactly that reason. In February 2026 Ipsos asked 970 adults in England and Wales about government plans to fund forty live facial recognition vans for town centres. Support ran at 64% against 17% opposed, and 53% said they would accept losing some privacy in public if it helped police catch criminals, against 24% who put privacy first.

So the fair summary is not "people object" or "people accept". It is that the answer moves twenty or thirty points depending on who is asking, who is operating, and whether the question mentions catching criminals. Any single number quoted on its own — including the ones in this article — is doing less work than it appears to.

The newest evidence is not a poll at all

Every figure above is someone answering a question. The strongest signal that sentiment is moving is not a survey response — it is communities removing cameras they have already paid for.

Flock Safety supplies automated licence plate readers to American police departments. Through 2025 and 2026 those contracts started going the other way. Newsweek's running tally, published on 28 August 2026, counts 149 cities, counties and local government bodies that have cancelled a Flock contract, rejected a proposed one, or switched off cameras already installed.

The reasons are specific rather than atmospheric:

• Lufkin, Texas. A police officer was indicted on felony charges after investigators alleged he accessed Flock data tens of thousands of times over a period of months.
• Pflugerville, Texas. The city voted to terminate its agreement after discovering that hundreds of outside organisations had run searches against data its own cameras collected.
• Stow, Massachusetts. Contract cancelled and cameras removed after an officer was accused of improperly accessing the data.
• Tempe, Arizona. System permanently deactivated. Officials said they had found no evidence of misuse locally, and cited growing concern about the technology anyway.

Tempe is the interesting one. A town that audited its own system, found nothing wrong, and switched it off regardless is not responding to a scandal. It is responding to a shift in what residents will tolerate — which is exactly the kind of change that a survey fielded in 2019 cannot see.

Note also what these cameras read. Flock's units capture licence plates, not faces. The objection they have provoked is not about the biometric specifically; it is about persistent, searchable records of ordinary movement, held by a private company, queryable by parties the community never approved. We have written separately on why plate readers do not need your face to track you and on how to find the ones near you.

What this data cannot tell you

Four things are worth stating plainly, because a chart makes everything look more certain than it is.

• The camera counts are estimates. They are assembled from procurement documents and reporting, and different researchers reach different totals for the same city. The ordering is more robust than any individual figure.
• Cameras are not facial recognition. A CCTV camera records. Whether that footage is run through face matching is a separate question, and camera density tells you nothing about it.
• Several of these surveys are from 2019, and they probably understate objection today. They predate Clearview AI's scraping becoming public, the EU AI Act, and the Flock cancellations below. Where older and newer evidence disagree here, the newer evidence points consistently one way: toward less tolerance, not more.
• Question wording moves these numbers more than anything else. "Is it acceptable" and "what do you expect will happen" produce very different answers from the same people in the same survey. Where a figure is quoted here, the question behind it is described rather than paraphrased.
• Survey conditions differ by country. Willingness to tell a stranger you object to state surveillance is not constant across the four countries in that study, and the authors note it. The China figure in particular should be read with that in mind.

Why any of this matters for your face

The public-space camera is the surveillance people picture, and it is the one they are polled about. It is also not the part that follows you around.

A camera on a street corner captures you in one place at one moment. A face-search engine does something categorically different: it indexes photographs that already exist of you — on social media, in news coverage, on a company's staff page, in someone else's holiday album — and makes them findable by your face alone. No camera involved. No consent asked. And unlike the camera on the corner, anyone with a browser can use it.

The Pew numbers suggest most people already sense this distinction, even if they have never had it put to them directly. Trust in police use runs at 56% and trust in advertisers at 18%, because people understand intuitively that accountability changes what a technology means. Face-search engines sit at the far end of that scale: commercial operators, no accountability to the person being indexed, and in most jurisdictions no obligation to tell them it happened.

That is the part of the problem an individual can actually do something about. You cannot opt out of a city's cameras. You can be removed from the databases that index your face.