Real-Time Facial Emotion Recognition System for Smart Classrooms

Abstract

The adaptive and personalized learning in the age of smart education requires real-time tracking of student interaction. Facial Emotion Recognition (FER) is a non-invasive and a high-performance tool in decoding emotional reactions of students in the classroom. This paper presents a lightweight, real-time FER system to be used in smart classroom environments, which takes advantage of deep convolutional neural networks (CNNs), which are optimized to run on the edge. The pipeline used in the system includes face detection, preprocessing, CNN-based emotion classifier, and dashboard visualization on live webcam feeds. It is tested on benchmark datasets (FER2013, RAF-DB) and on data recorded in the classroom and proved to be highly accurate on a per-class basis with low inference latency. With a mean frame rate of 25-59 FPS based on the hardware setup, the system is able to run continuously on any of the typical computing platforms. Also, analytics of emotion trends throughout the period of classes give useful feedback to teachers. The findings affirm the model to be suitable in the real time, ethical, and pedagogically effective classroom application.