Face recognition

During the course of the project, we created a face recognition system that can find an exact match in a database of more than 10,000 in 0.02 seconds. Due to the low resource intensity of the algorithm, we were able to place the algorithm on a single-board computer that is integrated with the camera

Facial recognition systems can be used for security in buildings, streets and public places.
Payment for goods and services
Such systems are used, for example, to pay for goods in stores, you can use facial recognition cameras to identify the buyer and deduct money from his account.

Access control
Facial identification systems can be used to control access to buildings, for example, in factories, factories and offices. Face recognition cameras can automatically open doors and control access to specific areas.

Main areas of application

Marketing and Advertising
These systems are also being developed to create personalized offers. For example, when recognizing the face of a customer in a store, you can offer him personal discounts and promotions.

Medical services
Face recognition algorithms can be used to identify patients and access medical history. For example, such a system can be used in hospitals and clinics to ensure security and control access to confidential information.

Tourism and hospitality
Customer check-in systems in hotels and airports can be considered as interesting cases. For example, by recognizing the face of a tourist at the airport, you can quickly complete all the necessary documents and reduce the waiting time.

The main difficulties of the task

It should be noted that the task of face recognition, despite the many options for its implementation, is associated with several aspects that often make it difficult to solve. Following are the main ones:
Differences in viewing angles
It is easiest to work with photos / videos taken from one angle. So, for example, a fairly common problem is the comparison of images of faces in front and profile, as well as at an angle.

Accessories and clothing details
Facial recognition systems can run into problems when trying to detect a person wearing glasses, a headgear, or a mask. As one of the well-known cases, one can recall how wearing masks during the coronavirus pandemic made it difficult to use the Face ID scanner to unlock Apple devices, this problem forced developers to make additional improvements for the convenience of users.

Image quality
The low quality and resolution of an image or frame can also cause an unsatisfactory result of the face recognition system, since in this case it is more difficult for the software to capture key points on a person’s face and, therefore, search for them.

Age changes
It is also worth noting that changing faces over time can also be a problem for face recognition systems, despite the fact that they are usually quite resistant to image searches in the range of +- 10 years if we are talking about adults.

As one of the examples of the implementation of a face recognition system, consider a case with an employee detector at the entrance of an enterprise.

It is assumed that such a system consists of two main components:
Single board computer installed with intelligent face recognition software
High-definition camera connected to it (from 1280x720)
An employee at the entrance to the workroom places his face in front of the camera, then the system searches the database for a face-to-face image from the video frame, and based on the search results, it concludes whether this person has access to the building.

The developed system is designed to reduce the burden on the enterprise security service, and also allows employees to enter the workplace without additional documents or electronic passes, which is also a convenience for employees.

Case: biometrics at the checkpoint of an enterprise

In less than a second, the system was able to find his image in the database by frame from the video, and also provided additional information about the distance (the smaller it is, the more likely it is that the video and the found image are the same person).

Also, the ability to hide irrelevant images at a given distance threshold was additionally configured. Thus, if the distance between images is greater than 0.51, it is considered that this person is not in the database, and the user is given information that the person was not found.

To  demonstrate the operation of the system, we developed a prototype that allows you to search for a person by  video.
 As an example, the results of a search video with message from former US President Donald Trump are shown.

As we can note, face recognition systems have a wide range of applications in a wide variety of areas. However, it is also necessary to additionally take into account such factors as the need for consent to the processing of personal data from users of facial recognition systems, as well as providing additional security measures when storing user data.


Facial recognition in store