课堂行为识别算法研究及智慧教室设计  

Classroom behavior recognition algorithm research and smart classroom design

作　　者：王萌 郑奋 WANG Meng;ZHENG Fen(Department of Computer and Simulation Technology,Faculty of Military Health Service,Naval Medical University,Shanghai 200433,China)

机构地区：[1]海军军医大学卫生勤务学系计算机与仿真技术教研室,上海200433

出　　处：《信息技术》2024年第11期132-139,共8页Information Technology

基　　金：2022年度海军军医大学卫生勤务学系教学研究与改革课题(2022WJA01)。

摘　　要：针对当前“一对多”的教学模式下,教师一人难以实时顾及全体学生状态的问题,提出一种多维度的课堂行为识别方法,实现对学生表情及动作的实时检测。该方法在运用卷积神经网络提取特征的基础上进行特征融合,提升了较小人脸表情识别的准确率;运用多帧识别、相似度比对、权重设置,提升了腿部遮挡动作识别的准确率。该方法已成功应用于学生表情及动作的实时检测,能有效完成对学生课堂行为的识别与分析,实验验证了该方法对学生行为识别的有效性。该方法可运用于智慧教室模型中,为教师掌握学生状态提供参考。In response to the current“one to many”teaching mode,which makes it difficult for a single teacher to take into account the real-time state of all students,a multi-dimensional classroom behavior recognition method is proposed to achieve real-time detection of student expressions and movements.This method uses convolutional neural network to extract features and then carries out feature fusion,which improves the accuracy of small facial expression recognition.By using multi frame recognition,similarity comparison,and weight setting,the accuracy of leg occlusion motion recognition has been improved.This method has been successfully applied to real-time detection of student expressions and actions,and can effectively identify and analyze student classroom behavior.The effectiveness of this method in student behavior recognition has been verified through experiments.This method can be applied to the smart classroom model,providing a reference for teachers to master student states.

关 键 词：表情识别 动作识别 智慧教室 深度学习 计算机视觉 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] G434[自动化与计算机技术—计算机科学与技术]