Short answer
A face recognition camera is a system that compares a face in the image with registered faces and determines who the person is. In organizations, it is used to open doors, to record staff entry, and to search for people in recordings. Because face data is biometric data, it counts as special-category personal data under KVKK; before installation, a legal basis, clear notice, and an alternative such as card access should be planned.
Face detection and face recognition are not the same thing
Two terms are often confused in camera catalogs. Face detection notices that there is a face in the image and saves a close-up of that face as a separate frame. It does not know who the person is. When you say ‘show me the faces that came through the door today’ in the recordings, it gives you a face gallery and speeds up review after an incident.
Face recognition goes one step further. It extracts a face template from the detected face and compares it with the templates of people registered beforehand. If it finds a match, it gives the person's name, opens the door, or raises an alarm. The legal and technical burden really begins at this second step.
The first question for your organization is this: do you need to know automatically who the person is, or is it enough to find faces quickly after an incident? For most security needs, face detection is enough and processes less data.
Where does it help in organizations?
Used in the right place and for the right purpose, face recognition provides real convenience. In the wrong place, it does not meet expectations and collects unnecessary data.
As you go through the list, ask the same question for each item: could a card, a password, or a plain camera recording do this job?
- Access control: staff passing through the door with a face reader terminal, without carrying a card.
- Staff attendance record: keeping entry and exit times with a face terminal.
- Unauthorized area alert: a notification to staff when someone who is not registered enters a restricted area such as a server room.
- Search in recordings: finding which cameras a particular person passed during the day.
- Visitor management: recognizing a pre-registered visitor at the entrance.
What determines accuracy?
Face recognition gives its best results in a narrow passage that is well lit, where the person faces the camera. A door entrance, turnstile area, or narrowed corridor is suitable in this respect. Expecting face recognition from a camera looking down from the ceiling in a wide lobby leads to disappointment.
Position the camera close to eye level, facing the direction the person is walking. If the camera is mounted too high, it sees nothing but the forehead and hair. A camera looking from the side sees half the face. If daylight comes in from outside the door, backlight darkens the face; a camera with wide dynamic range or extra lighting is needed.
Masks, hats, thick-framed glasses, and hair covering part of the face lower the match rate. The quality of the registered photo also matters; a blurry or side-angle registration photo gives wrong results even on a good camera.
Expert note: Run a face recognition installation in trial mode for a few weeks. During this time, see how many people the system fails to recognize and whom it wrongly matches, and adjust the camera angle and match threshold accordingly.
False matches and human confirmation
No face recognition system is error-free. There are two kinds of errors: the system fails to recognize a registered person, or it confuses one person with another. In actions such as opening a door, the first error keeps the person waiting at the door; the second error can let an unauthorized person in.
That is why the match threshold is set according to the security level. At a sensitive door such as a server room, the threshold is kept high and a second verification such as a card or password is added next to the face. When an alert flags someone as ‘suspicious’, the decision should be left to a staff member, not to the system. Taking action against someone based on an automatic match has serious consequences in the case of a false match.
KVKK: why does biometric data need extra care?
Law No. 6698 on the Protection of Personal Data counts biometric data among special-category personal data. The face template extracted for face recognition falls within this scope. The conditions for processing special-category data are stricter than for ordinary camera recordings.
In practice, this means: you need to determine on what legal basis you are processing, inform people clearly, take additional measures for the security of the data, and define the retention period. For things like staff entry records, offering the employee an alternative such as card access is the path many organizations prefer. Be especially careful in places where children are present, such as schools.
This is not a technical decision but a legal one. Before installation, assess the process with a lawyer or a KVKK consultant and put the decision in writing.
What can be used instead of face recognition?
The need that makes you think of face recognition can often be met with a simpler method. If the goal is to keep unauthorized people from passing through the door, card access or a keypad with a code does the same job without processing biometric data. If you are worried about a card being given to someone else, a camera facing the door records who swiped the card; that recording is checked after an incident.
If the goal is to find a person quickly in recordings after an incident, face detection and the recorder's face search feature are enough in most cases. Instantly recognizing certain people in a crowded area and raising an alert is the hardest use, both technically and legally; if you are starting from this need, do not rush the decision.
There are also devices that offer card access, password, and a face terminal together on the same door. That way, an employee who does not want their face registered passes with a card, and the system collects both methods in the same record.
Decisions to make before installation
Before installing a face recognition system, answer a few questions within your organization. At which doors will it be used? Whose faces will be registered, and how will this registration be done? When staff leave, who will delete the face template, and when? How will people get through the door when the system is not working?
On the technical side, ask where the face data is kept: inside the terminal, on the organization's server, or on an internet service? Whether the data leaves the organization matters for both security and KVKK. Keeping the terminals and the recording server on a separate network, changing administrator passwords, and keeping up with software updates are also part of the installation.
Frequently asked
Does a face recognition camera recognize someone wearing a mask?
Some systems have been developed to also work on a masked face, but masks, hats, and glasses lower the match rate. The exact result is seen during a trial period, depending on the model used and the installation conditions.
Is it legal to track staff entry and exit with face recognition?
Because face data is biometric data, it counts as special-category personal data under KVKK and its processing is subject to stricter conditions. Consult a lawyer before installation on the legal basis, notice, and alternative methods.
Can face recognition be added to my existing cameras?
Some recorders and software can do face recognition on the image of existing IP cameras. However, if the camera's angle, height, and resolution are not suitable, the result will be poor. Nothing definite can be said without seeing where the existing cameras are.
The right solution for your organization
Tell us what you need, and Desa Bilişim will assess face recognition, face detection, or card access options on site and prepare a written quote.
