作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程 ›› 2024, Vol. 50 ›› Issue (7): 96-103. doi: 10.19678/j.issn.1000-3428.0069592

• 智慧教育 • 上一篇    下一篇

基于全范围头部姿态估计的教师注意力识别算法

陈增照*(), 王政, 郑秋雨   

  1. 华中师范大学教育大数据应用技术国家工程实验室, 湖北 武汉 430079
  • 收稿日期:2024-03-18 出版日期:2024-07-15 发布日期:2024-07-09
  • 通讯作者: 陈增照
  • 基金资助:
    国家自然科学基金(62077022); 华中师范大学国家教师发展协同创新实验基地建设项目(CCNUTEIII 2024-01)

Teacher Attention Recognition Algorithm Based on Full-Range Head Pose Estimation

Zengzhao CHEN*(), Zheng WANG, Qiuyu ZHENG   

  1. National Engineering Research Center of Educational Big Data, Central China Normal University, Wuhan 430079, Hubei, China
  • Received:2024-03-18 Online:2024-07-15 Published:2024-07-09
  • Contact: Zengzhao CHEN

摘要:

探究教师注意力对于评估课堂教师行为具有极其重要的研究价值。然而, 现有的教师注意力识别算法存在无法应对极端头部姿态角度等问题。为此, 提出一种基于6DRepNet360模型的教师注意力状态识别算法, 提升极端角度中头部姿态估计算法的准确性。相较于传统的依赖条件判断来分类教师注意力状态的方法, 设计一种基于支持向量机(SVM)的教师注意力分类模型, 对复杂头部姿态角度进行注意力状态的精准识别。为进一步解决算法稳定性和准确性带来的误差数据, 提出基于滑动窗口的数据清洗算法, 有效提高整体识别结果的真实性和可靠性。通过在构建的CCNUTeacherState数据集上进行一系列的算法评估, 实验结果表明, 所提出的教师注意力识别算法在CCNUTeacherState数据集上达到了90.67%的准确率。

关键词: 教师注意力, 全范围角度, 6DRepNet360模型, 支持向量机, 数据清洗技术

Abstract:

Research on teacher attention is valuable for evaluating classroom teacher behavior. However, existing teacher-attention recognition algorithms struggle to address issues such as extreme head pose angles. This study proposes a teacher-attention state recognition algorithm based on the 6DRepNet360 model. This algorithm effectively enhances the accuracy of head pose estimation algorithms under extreme angles. In contrast to traditional methods that rely on conditional judgments to classify teacher attention states, the designed teacher-attention classification model, based on a Support Vector Machine(SVM), enables precise recognition of attention states even under complex head pose angles. To further address the errors introduced by the algorithm stability and accuracy, this study introduces a data cleaning algorithm based on a sliding window, effectively improving the authenticity and reliability of the overall recognition results. Through a series of algorithm evaluations on the constructed CCNUTeacherState dataset, experimental results demonstrate that the proposed teacher-attention recognition algorithm achieves an accuracy of 90.67% on the CCNUTeacherState dataset.

Key words: teacher attention, full-range angles, 6DRepNet360 model, Support Vector Machine(SVM), data cleaning technology