[D] Real-time Student Attention Detection: ResNet vs Facial Landmarks - Which approach for resource-constrained deployment?

A new biometric analysis published in Frontiers in Computer Science offers a breakthrough for real-time student attention detection in educational settings. The research team conducted eye-tracking experiments with 30 participants to scientifically determine which facial regions humans actually use when recognizing emotions. Their key finding reveals that people focus primarily on the eyes (with particular emphasis on the left eye) and the mouth, rather than processing the entire face uniformly. This discovery enabled researchers to innovate by reducing the standard 68-point facial landmark model down to just 24 critical points that capture these essential features.

This streamlined approach presents a compelling alternative to deep learning models like ResNet for resource-constrained deployment scenarios. While ResNet models process raw facial images through convolutional neural networks to output emotion classifications, they require significant computational resources. The facial landmarks method, based on geometric measurements of specific coordinate points rather than pixel patterns, offers dramatically reduced computational overhead. For classroom applications where real-time processing on limited hardware is essential, this 24-point model provides a practical solution for detecting student engagement levels—identifying whether students are engaged, confused, or bored—without the heavy infrastructure requirements of full deep learning implementations.