Research on Learning Status Evaluation in the Classroom Based on End-to-End Head Pose Estimation

In the context of AI-enabled education, this paper seeks to improve the accuracy of evaluating learning statuses in classroom through the design and improvement of deep learning models, as well as the evaluation methods of students' status and attention in classroom. Initially employing deep learning techniques to analyze students' head poses, this study addresses the challenge of estimating head poses in natural teaching environments where facial image details are often insufficient. The research explores a fine-grained head pose estimation method that does not rely on facial landmarks. By integrating classification and regression within a multi-loss network, the study predicts the Euler angles of head poses from students' facial images, thereby achieving end-to-end head pose estimation. Subsequently, in the evaluation method of student statuses and attention in classroom, we developed a head state evaluation method suitable for this head pose estimation model within the smart classroom environment at Central, and designed a diversified evaluation method based on the currently widely used head pose evaluation method of student attention. A comprehensive and effective assessment of student attention in classroom is carried out. According to experiments, results demonstrated the effectiveness of the proposed diversified evaluation methods, which outperform singular evaluation approaches in terms of performance. This method enables teachers accurately grasp the status of students in the classroom, so as to adjust the teaching rhythm, carry out teaching reflection, and also contribute to teaching evaluation.