Do Security Cameras Have Facial Recognition?

The short answer

Across every camera and doorbell in our registry, we found no first-party claim of facial recognition from eufy, Reolink, TP-Link, Wyze or Blink. What those makers publish instead is detection by object class: person, pet, vehicle, package. Useful, and a different thing entirely.

We record that as “not published” rather than “does not exist”, because these brands sell more products than we cover and feature lists change. But if you are shopping with the expectation that a home camera will tell you who is at the door rather than what is at the door, the published evidence does not support it.

The two features people confuse

Person detection answers one question: is that a human? It separates a person from a cat, a passing car, rain on the lens or a branch moving in the wind. It does not know who the person is, and it keeps no record of their face. Its entire job is reducing the number of useless alerts on your phone, which is the actual daily problem with owning a camera.

Facial recognition answers a different question: which human? To do that, something has to build a template from a face and store it so later faces can be matched against it. That is biometric data about everybody who walks up to your door, including couriers, neighbours and children, and it carries privacy, security and legal weight that ordinary motion video does not. Federal bodies publish consumer guidance on home camera privacy and account security for a reason — the FTC on securing cameras, NIST through its IoT cybersecurity programme — and a face database at the front door raises the stakes on every one of those recommendations.

Most people asking for facial recognition want the first thing. They want the camera to stop pinging them about the wind.

What each brand actually publishes

Reolink is the most specific. Its 2nd-Gen battery doorbell lists person, vehicle, animal and package detection on the product itself, and Home Assistant’s Reolink integration turns three of those into binary sensors — AI person, vehicle and pet detection — alongside motion and doorbell-press sensors. No identity, four classes, and no monthly fee: Reolink publishes “No Monthly Fees for Local Storage” with microSD, Home Hub, NVR and FTP as the destinations.

TP-Link matches the class list and is explicit about the price. For the Tapo D225 it lists motion, person, pet, vehicle and package detection as free AI detection, with Tapo Care cloud storage as the separate paid extra. Its own Kasa FAQ goes further, putting live view, instant notifications, two-way audio, activity zones, local storage, motion detection and person detection in the no-subscription column of its feature table.

eufy approaches it from the hardware side rather than the software side: the eufyCam S330 publishes radar plus PIR detection, which is two different physical sensors agreeing that something is actually there. That is a genuinely effective way to cut false alerts, and it says nothing about identity.

Ring is where the line is sharpest, and it is a pricing line rather than a technical one. Ring’s plans page lists person, package and vehicle alerts as plan features, and its own Alexa article states that doorbell press and motion announcements can be enabled without a plan while person and package announcements require a Ring Protect plan. So on Ring, even telling a person from a car is the paid tier.

Detection, by brand and by price

Two readings of that table. First, nobody in this price bracket is selling identity — so a roundup promising “the best facial recognition cameras” is either covering products we do not carry or stretching the words. Second, the brands differ enormously on whether the useful version of this feature costs money, and that is the row to shop on.

Why face matching is harder than it sounds at a front door

Even where face recognition exists, the physics of a doorbell work against it. Night vision on these products is mostly infrared: Reolink publishes 850nm IR LEDs, Wyze publishes four 850nm IR LEDs plus two spotlight LEDs, and Blink describes 850nm infrared producing black-and-white video with a glare warning for reflective surfaces. Infrared throws away the colour and subtle tonal detail that face matching leans on — so the half of the day you most want to identify somebody is the half where the camera is least able to.

Add the angle. A doorbell mounted at chest height looks up at a hood, a cap and a parcel held in front of a chin. Add weather, backlight at dusk and the three seconds a courier spends on your step. Class detection copes with all of that fine. Identity does not.

If seeing faces clearly is the real goal, buy for light rather than for AI: a spotlight model that switches to full colour at night, as covered in our night vision ranking, and placement that puts the face in frame, which is the subject of where to place security cameras.

The three we would buy for smarter alerts

These picks come from our main camera ranking and doorbell ranking, chosen on the published detection classes and storage in the table above rather than on testing we have not done.

If you genuinely want identity, do it locally

There is an honest route to face matching at home, and it is not a consumer camera feature: run the recognition on hardware you own. People do this with Home Assistant and local image-processing projects, feeding it from a camera that exposes a local stream — which is exactly what Home Assistant’s Reolink integration provides, and what TP-Link’s RTSP route provides on its hardwired models.

We are not going to rank products on community add-ons, so we will not name a project as a pick. What we will say is that the architecture is the right one: if a face template is going to exist, it should live on your server rather than in a company’s cloud, and our Home Assistant camera picks and privacy-first hub ranking are where that build starts. Before you do any of it, read how to secure smart home devices, because a face database with a reused password is worse than no face database at all.