一种用于“远程课堂”的学生听课专注度自动评估方法  

Automatic Concentration Assessment for Student in “Remote Classroom”

作　　者：邵帅 李思敏 广田薰 戴亚平[1,3] SHAO Shuai;LI Simin;HIROTA Kaoru;DAI Yaping(School of Automation,Beijing Institute of Technology,Beijing 100081,China;The Fiftieth Research Institute of China Electronic Technology Group Corporation,Shanghai 200331,China;Tangshan Research Institute,Beijing Institute of Technology,Tangshan,Hebei 063611,China)

机构地区：[1]北京理工大学自动化学院,北京100081 [2]中国电子科技集团公司第五十研究所,上海200331 [3]北京理工大学唐山研究所,河北唐山063611

出　　处：《北京理工大学学报》2024年第5期530-537,共8页Transactions of Beijing Institute of Technology

基　　金：北京市高等教育学会2022年面上课题(MS2022403);国家自然科学基金资助项目(82201753)。

摘　　要：在基于互联网的“远程课堂”中,如何利用学生课堂状态视频数据建立教师与学生授课−听课之间的评价关联关系,是一项富有挑战性的科学问题.本文针对“远程课堂”中的学生行为检测识别问题,提出针对课堂学生“面部姿态角度”检测模块与“身体动作行为”识别模块,对学生的课堂行为进行识别分类;针对“学生听课专注度”的定量分析问题,提出一种基于学生面部姿态角度和行为分类结果的量化评估算法;运用证据理论对“面部姿态”与“动作行为”并行进行数据融合计算,建立用于“远程课堂”的在线“学生听课专注度”自动评估系统模型.本文所提模型能够对课堂学生的听课行为进行在线检测与分析,完成“学生听课专注度”的定量评分并输出评估结果.实验中,系统对“学生听课专注度”的评估准确度达到90.4%,验证了系统的有效性.In the internet-based'remote classroom',it is a challenging scientific problem to establish an evaluation correlation for the lecture-listening state between teacher and students based on video data of student classroom status.In this paper,a module was proposed firstly to identify and classify the classroom behaviors of students according to their'facial posture angle'and'body movement behavior'in'remote classroom'.And then,a quantitative evaluation algorithm was proposed based on student facial gesture angle and behavior classification results to analyze quantitatively the student attentiveness.Finally,using evidence theory to carry out the data fusion for student facial gestures and behavioral classification results in parallel,an automatic assessment system model was established to analyze automatically the student on-line concentration in remote classroom.The results show that the proposed model can detect and analyze student listening behaviors,complete score quantitatively and output the evaluation results for student concentration.In concentration assessment experiments,the accuracy of the system can reach 90.4%,verifying the effectiveness of the system.

关 键 词：深度学习 图像分析 行为识别 数据融合 远程课堂 专注度评估 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]