EduNet: A New Video Dataset for Understanding Human Activity in the Classroom Environment

Area of Science:

• Computer Vision
• Artificial Intelligence
• Educational Technology

Background:

• Human action recognition (HAR) research has advanced significantly in various domains like sports and daily activities.
• Existing HAR datasets lack focus on educational environments, hindering AI development in this area.
• Classroom monitoring systems require specialized datasets for effective teacher and student activity recognition.

Purpose of the Study:

• To introduce EduNet, the first dataset specifically designed for human action recognition in classroom environments.
• To address the research gap in HAR datasets for educational settings.
• To facilitate the development of advanced AI-powered classroom monitoring systems.

Main Methods:

• Developed EduNet, a novel dataset comprising 7851 manually annotated video clips.
• EduNet features 20 distinct action classes, covering both teacher and student activities.
• Clips were sourced from YouTube and recorded in actual classroom settings, totaling approximately 12 hours.

Main Results:

• EduNet is the first dataset for classroom monitoring of both teacher and student activities.
• The dataset presents a challenge due to its large number of unconstrained clips.
• An I3D-ResNet-50 model achieved 72.3% accuracy on EduNet, outperforming benchmarks like UCF101 and HMDB51.

Conclusions:

• The EduNet dataset is a valuable resource for advancing HAR research in education.
• This benchmark dataset will accelerate the development of intelligent classroom monitoring solutions.
• Future research in educational AI can leverage EduNet for improved system performance and insights.