AI and ML-based solutions
R&D. Development. Consulting.
Module for recognizing faces of videoconference participants
HOME / OUR WORK / Recognizer
/ task

Create a module that can recognize in real time in the video stream the faces of video conference participants, give them access to the conference by face, and put the appropriate signature of the full name for each participant.

/ Solution
The service works tightly with the API of the videoconferencing server, uses its video stream without overloading the main server and recognizes faces with high accuracy.
We also realized seamless integration of the functionality into an existing high-loaded product through close cooperation with the in-house development team.
/ technologies
Technological complexity: smart search, which takes into account additional parameters when ranking similar faces of employees; development of its own workflow of a cascade of neural networks and machine learning algorithms to quickly find matches for all faces in the frame, detection and 3D reconstruction for image alignment and verification of permissible limits for triggering facial rotation recognition (increased accuracy, reduced errors). Similar approach is applied in their products by market leaders specializing in face recognition and identification (e.g., VisionLabs).