AI Powered Eye Tracking for Attention Analysis in Classroom Videos

Author : Priyansh Pal

Abstract :Effective teaching depends on student attention; but, manual evaluation presents major difficulties. This work presents an artificial intelligence-driven eye tracking system using remote gaze estimate for classroom video analysis. Integrating facial features, eye region details, and temporal cues, this multi- stream deep neural network estimates three-dimensional gaze direction and classifies attention. This work reports a dual- task loss that simultaneously maximises attention classification and gaze error. Evaluations on standard benchmarks (MPIigaze, Gaze360, TabletGaze) and real classroom data show a mean angular gaze error of less than 3.5° for frontal views and approximately 10° in wide-angle settings, so attaining over 90% accuracy in attention classification, surpassing previous methods. This research improves automated classroom analytics meant to support tailored learning and offer instructional comments.

Keywords :Eye Tracking, Attention Analysis, Convolutional Neural Network, LSTM, Classroom Videos.

Conference Name :International Conference on Engineering & Technology (ICET-25)

Conference Place Delhi, India

Conference Date 26th Apr 2025