Architecture of a Real-Time Facial Emotion Recognition System for Intelligent Systems

Abstract

The study aimed to design and evaluate the architecture of a real-time facial emotion recognition system for intelligent applications integrated into the educational infrastructure of Akhmet Baitursynuly Kostanay Regional University (Republic of Kazakhstan). The paper analyzes modern machine learning approaches for computer vision, including convolutional neural networks and transformer architectures, to identify and compare baseline models for automated visual information analysis. Experimental evaluation was conducted using the publicly available Facial Expression Recognition 2013 (FER-2013) and Real-world Affective Faces Database (RAF-DB) datasets, which are widely used for benchmarking emotion recognition methods. The proposed architecture achieved higher emotion recognition accuracy than the baseline models while maintaining comparable computational costs and improving video processing efficiency. It also required fewer computational resources, particularly in terms of memory consumption and graphics processing unit load. The developed system demonstrated stable performance during real-time video stream processing and consistent recognition accuracy across different test datasets. In addition, the model maintained reliable emotion classification under variations in input images throughout the experiments. The practical significance of the research lies in the applicability of the proposed system to real-world video analysis tasks, including educational, research, and intelligent applications requiring automatic real-time recognition of users’ emotional states under limited computational resources. The findings can support the development of intelligent systems for behavioral analysis and the investigation of human interaction with digital environments, providing an efficient and resource-conscious solution for real-time emotion recognition.