RSS
Логотип
Баннер в шапке 1
Баннер в шапке 2

MISIS: Facial Recognition System

Product
Developers: Mining Institute MISiS (former Moscow State Mining University, Moscow State Mining University), Higher School of Economics (HSE), Sberbank
Date of the premiere of the system: December 2023
Branches: Information Technology
Technology: Video Analytics Systems

2023: Technology Announcement

MISIS, HSE and "" Sber created a technology that is 30 times faster than others recognizes faces in the video. This development became known on December 8, 2023.

Its developers told Izvestia about the principles of the algorithm. According to them, first two frames are analyzed - the first and the last. If AI can draw the right conclusion, then its work stops. Otherwise, a few more personnel is added in the middle, and then the correct pins are checked again. For example, for simple videos, where facial expressions are well recognized, the decision is made almost instantly, and for more complex ones, much more video frames must already be processed.

Created a technology that emasculates faces on video 30 times faster than others

As noted by Andrei Savchenko, Deputy Director of the Center for Artificial Intelligence at NITU MISIS, to optimize the method, it is important not only to collect information from all video personnel, but also to be able to correctly build a decision-making sequence, that is, to immediately weed out unnecessary information so that only the data necessary for classification remain.

File:Aquote1.png
Our approach is based on sequential analysis of input video data with different detail (frame rate). This method is designed to be very fast and applied to any way of extracting characteristic features of faces based on deep neural networks without additional training of the entire model, Savchenko explained in a conversation with the publication.
File:Aquote2.png

The authors of the projects say that the method can be used for video analytics on any hardware that does not require computing resources: for example, on a budget smartphone, laptop, smart camera, and so on. At the same time, there is no need to send video to a remote computing server - this increases the level of protection of personal data, added Andrei Savchenko.[1]

Notes