The NeuroVision neural network looks at a frame and returns a person’s approximate age, gender and emotion — and also whether they wear glasses or a mask and where the head is turned. The face is not compared with a database and not recognised: the response contains no name, only an estimate. This is how the prompt for a cashier at an 18+ checkout, an anonymised audience report and an age check from a selfie against a document work.
Face analysis
An age and emotion estimate is a probabilistic result, not a document and not an identification of a person
The API response is “≈ 20–26 years, female, joy”, not “Camilla, a regular customer”.
Face recognition usually answers the question “who is this?”. Attribute analysis answers a different one: “what kind of person is in front of the camera right now?” — roughly how old they are, whether they are a man or a woman, whether they are smiling. This needs neither a face database nor a name. A camera or an app sends a frame, and the API returns the facial attributes and the confidence for each. The next frame is analysed anew: the system does not remember having seen this person before and does not link two frames together.
Age comes as a range, emotion as a probability. It is a hint for a decision made by a person or by your rule: call a salesperson, show a different video, ask to retake the selfie. For the decision “sell or not” the law requires a document.
Six attributes determined by NeuroVision face analysis. You can enable one — for example, only age at the checkout — or all of them at once for an audience report.
Age detection from a face is an estimate as a range of years. It suits understanding whether a customer looks like an adult with a margin, and age groups in a report: under 25, 25–34, 35–44 and older.
Gender detection from a face is male or female with a confidence value. In a KYC session the gender from the selfie is matched with the gender in the document: a mismatch is a reason to look at the session more closely.
The facial expression in the frame: joy, neutral, surprise, sadness, irritation and others. It is used in audience reports, not for assessing a person.
Whether the face has glasses, a medical mask or a beard. It helps understand why an estimate is uncertain and ask to take off the mask before the selfie.
The tilt and rotation of the head in degrees. If a person has turned away, the age and emotion estimate cannot be trusted — it is better to skip the frame.
The liveness check tells a person apart from a photo, a phone screen and a mask. More on the page Liveness.
An example is a self-checkout at a neighbourhood store. A customer scans a bottle of wine, and the checkout asks her to look into the camera.
The checkout camera picks out the face in the frame and checks whether it can be judged: the face is large, looks into the camera and is not covered by a mask. If not — the checkout asks the person to step closer.
Before estimating age, the system checks that a live person is in front of the camera, and not a photo of an adult on a phone screen.
The neural network estimates age as a range. You set the threshold with a margin, for example “younger than 25 — a document is needed”, so that an adult who looks young does not pass without a check, and not the other way round.
If the estimate is below the threshold or uncertain, the checkout calls a salesperson. The sale is allowed by the salesperson after seeing a document or data from the information exchange service — as the law requires. The frame is not stored after the response.
What the law requires
If there is doubt about majority, an alcohol seller has the right to ask for a document or to check the age data through the multifunctional information exchange service. A seller of cigarettes, vapes and hookahs is in this case obliged to ask for a document and to refuse if there is none.
Where the camera helps
A customer who looks clearly adult with a large margin passes without waiting. An uncertain estimate or one below the threshold is a signal for the salesperson to come over. Fewer “just in case” calls — a shorter queue at the checkout.
Where the camera does not decide
The estimate does not confirm age and does not replace a document. If the salesperson is in doubt, the decision is made by the document alone. Set the threshold with a margin: letting a teenager through costs more than asking for a passport once too often.
What the salesperson sees
The screen shows “age check needed” and the product. Not a name, not a photo from a database and not a purchase history: the system does not recognise the customer and will estimate anew next time.
A face-based age estimate does not make the sale of alcohol or cigarettes legal by itself. Its job is for the salesperson to approach those who really need to show a document, and not everyone in a row.
Which goods cannot be sold to minors, what fines threaten the seller and how to confirm majority from a document during online registration — on the page “Age verification 18+”.
Gender, age and emotions are counted from camera frames and added up into a report: how many people of which group came up and how they reacted. The report holds numbers and shares, not people.
Visitor counting, heat maps and queues — on the page video analytics.
A camera next to a digital signage screen shows who watches the video: age group, gender, facial expression at the moment of display. You can compare two videos on the same audience.
Suitable for:
Shopping malls, chains, petrol stations
How many people lingered at the stand and who they are by gender and age. It helps understand who the promotion really works for.
Suitable for:
retail, exhibitions
Shares of neutral, satisfied and irritated faces at the counter at different hours. A signal to look at the queue and the shift load, not an assessment of a specific employee.
Suitable for:
banks, public service centres, service centres
The reaction of research participants to a mock-up or a video — with their consent, in a testing room or via a link on a smartphone.
Suitable for:
research, UX
Not a lie detector
A nervous smile, tiredness, and individual and cultural features of facial expression produce the same look as an untruth. You cannot decide from an emotion whether a person is telling the truth — and our API is not designed for that.
Not an assessment of trustworthiness
An emotion in the frame does not show whether a client or a candidate is honest. To check a client there are a document, Liveness and database checks — on the KYC platform, not facial expressions.
Not an assessment of employees
We do not suggest rating employees and candidates by facial expression. A service report contains shares by zone and time, with no link to a specific person.
Not a diagnosis and not recognition
An emotion is not a medical conclusion about a condition. And attribute analysis does not recognise a person: if the task is to find a specific person in your database, this is face search with its own rules.
The neural network sees a facial expression — a smile, furrowed brows, an open mouth. It does not see what a person thinks and is going to do. That is why we state plainly what emotion detection is not suited for.
Where emotions are useful: to compare the audience’s reaction to two videos, to see at which hours there are more irritated faces at the counter, to check whether an interface is clear to research participants. Everywhere these are shares for a group, not a conclusion about a person.
Attribute analysis is built so as not to establish identity: which legal regime applies to it depends on that. Below is a summary of the norms as of October 2026. It is a guide for a conversation with a lawyer, not a legal opinion.
Where it matters what audience is in front of you, not who exactly they are.
Frame attributes are enabled in the personal account and via the API: the ones you need — by role. Get a test on your own frames
A free test on your own frames, a cloud in Russia or your own server, payment per check.
The price depends on where the attributes are calculated: in a KYC session, on separate frames via the API or on a camera stream.
Document, selfie and Liveness; the age and gender check from the face against the document is included in the “Document + Liveness” package.
on request
Separate frames from a checkout, an app or a test bench; only the attributes you need.
on request
A camera stream in a hall, a report by group, installation on your servers.
KYC package prices — in the price matrix. Equipment, cameras and installation are not included in the price.
Terms you will meet in the project, the contract and a conversation with a lawyer.
The project «Development of a multimodal system using artificial intelligence "Neuromed" for managing the diagnostic processes of a clinic on the basis of the "Neurovision" computer vision platform» is supported by the Innovation Promotion Foundation within the federal project «Domestic Solutions»
The Neurovision computer vision platform is registered with Rospatent (certificate No. 2023661508) and included in the Russian Software Registry (entry No. 19972 of 13.11.2023)
The right to use the Platform is granted under a licence agreement concluded by Neurovision with its customers
Tell us where you need the analysis — a checkout, screens, a hall or online registration — and which attributes matter. We will show the result on your own frames.
Can't put your request into words? Book a call and our team will contact you at a convenient time
Book a callThe company is included in the register of IT companies and the register of personal data operators of the Russian Ministry of Digital Development.
Address: Moscow, Presnenskaya Embankment 12 — Moscow City business centre, Federation Tower. Nearest metro stations: Delovoy Tsentr and Mezhdunarodnaya.
Articles on face recognition, liveness checks and the protection of biometric data — in our blog.
The neural network finds the face in the frame, checks that it is a live person and estimates age as a range of years. The face is not compared with a database, so the response contains no name — only an estimate and a confidence value.
No. The estimate hints who should show a document. If in doubt, an alcohol seller has the right, and a tobacco seller is obliged, to ask for a document or data through the multifunctional information exchange service. More on the page “Age verification 18+”.
It is an estimate, not an exact figure: it is affected by light, angle, make-up, a mask and individual features. That is why the threshold for a checkout is set with a margin, and in case of doubt the document decides. You can check the accuracy on your own stream in a free test on your own frames.
The law calls biometric data the information by which identity is established and which is used for that. Attribute analysis does not compare the face with a database and is not used to establish who the person is. At the same time, a frame with a face is personal data and must be handled under 152‑FZ.
The facial expression at the moment of the frame: joy, neutral, surprise and others. This is useful for a report on the audience’s reaction, but it does not show whether a person is telling the truth and is not suitable for assessing employees or candidates.
No. Each frame is analysed anew, and no template for search is stored. Searching for a person in your own database is a separate scenario with its own 572‑FZ rules: face search.
Frames are not needed for the audience report: numbers by group go into the report. Storing frames in a KYC session is set in the scenario. It works in the cloud on servers in Russia or on your own servers.
No. It is a special category of personal data, and it is not needed for the tasks of this page.
For a selfie — the front camera of a smartphone or a laptop. For a checkout and a hall — a camera that sees the face large and from the front, without backlight. We will choose one for your site.
The age and gender check from the selfie against the document is included in the “Document + Liveness” KYC session from 29 ₽. The API for separate frames and a camera stream — on request.
