Back to search

Institutional Repository of Vilnius University

Face recognition in communication with a robot /

Abstract

dc:description

This paper presents the development of an open-source Python-based real-time face recognition system designed for efficient and dynamic face recognition applications that includes face tracking logic, particularly in robotic contexts. The system integrates multiple face detection and encoding components as well as a tracking system to target different use cases. Key contributions include the implementation of a full end-to-end face recognition pipeline that uses a combination of MediaPipe, YUNET, and RetinaFace for face detection, allowing adaptation based on specific use-case requirements for speed or accuracy. Face encoding is performed using two variants of the FaceNet model: FaceNet 128 for fast real-time performance and FaceNet 512 for higher accuracy. To improve tracking accuracy and stability, the system employs a Unified Kalman filter with an Intersection Over Union (IOU) metric introducing SORT-N, Stable Online Realtime Tracking with Nonlinear Dynamics, which is a modification of the regular SORT algorithm that proposes a way to improve the real-time dynamic face tracking by taking into account non-linear motions and rapid changes in face sizes in the recordings. The developed system also includes optional encryption for secure embedding management, using SQLite as a persistent storage mechanism integrated with HNSWlib for efficient in-memory searches. During testing on videos, the fastest pipeline configuration, which used MediaPipe for detection and FaceNet 128 for encoding, achieved 79.77 FPS when processing every single frame. When using face detection interval value equal to 5, face detection was performed on every 5th frame with face tracking utilized for the frames in between, achieving 166.7 FPS. The pipeline that provides the optimal combination of accuracy and speed, which uses YUNET for detection and FaceNet 128 for encoding, achieved 48.49 FPS with face detection interval of 5.

Degree

thesis:*
Grantor dc:publisher
Institutional Repository of Vilnius University
Year dc:date
2025

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Iachnyk, Ivan,

Subjects

dc:subject × 1

Rights

dc:rights
Statement dc:rights
  • info:eu-repo/semantics/openAccess
Language dc:language
eng

Identifiers

dc:identifier.*

Chain of custody

source
Harvested from
Vilnius University
Base URL
epublications.vu.lt/oai
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
citation

Iachnyk, Ivan,. Face recognition in communication with a robot /. Institutional Repository of Vilnius University, 2025. https://repository.vu.lt/VU:ELABAETD229558604&prefLang=en_US