Student Posture Extraction and Attentiveness Classification Using Deeplabcut and Support Vector Machine

This study proposes a computer vision system to classify student attentiveness in physical classrooms using upperbody posture analysis. Leveraging the DeepLabCut framework for pose estimation and a Support Vector Machine (SVM) classifier, the system processes video data captured via a Raspberry Pi 5 and an OBSBot Tail Air camera to categorize attentiveness into three levels: fully attentive, partially attentive, and nonattentive. Unlike existing methods focused on online learning environments, this approach targets physical classroom settings, addressing gaps in engagement monitoring. The model was trained on annotated posture data from volunteers, with body part coordinates normalized relative to the thorax for consistency. Evaluation using a confusion matrix revealed an accuracy of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{7 0. 8 5 \%}$</tex>, highlighting challenges in distinguishing subtle posture variations between attentiveness levels. The system provides instructors with actionable insights to adapt teaching strategies, promoting personalized learning. This work demonstrates the feasibility of integrating low-cost hardware and machine learning for classroom engagement analytics, though future studies could benefit from larger datasets and enhanced classification models.