基于深度学习的课堂注意力检测  

Classroom attention detection based on deep learning

作　　者：于子平 程杰 丁岩康 仇睿 王单单 李伟仙[1] Yu Ziping;Cheng Jie;Ding Yankang;Qiu Rui;Wang Dandan;Li Weixian(School of Instrumentation Science and Optoelectronic Engineering,Beijing Information Science and Technology University,Beijing,100192)

机构地区：[1]北京信息科技大学仪器科学与光电工程学院,北京100192

出　　处：《电子测试》2021年第22期55-57,共3页Electronic Test

基　　金：北京信息科技大学科研类校内项目重点研究培育项目(2121YJPY206);北京信息科技大学研究生课程建设项目(2020YKJ08)。

摘　　要：为了无介入的量化学生上课的精力集中程度,提出并设计了基于深度学习的课堂注意力检测算法。利用libfacedetection人脸检测模型检测摄像头画面中含68个特征点的人脸,再根据特定的6个特征点估计头部姿态以及眼部12个特征点计算疲劳程度,并通过特征融合的策略获取学生的注意力状态,从而计算学生注意力集中时间。实验结果表明,该算法具有较高的准确率,达到了预期效果。In order to quantify students’concentration degree in class without intervention,a class attention detection algorithm based on deep learning is proposed and designed.The libfacedetection face detection model is used to detect the faces with 68 feature points in the camera image.Then,the head posture is estimated according to the specific 6 feature points and the fatigue degree is calculated with 12 eye feature points.At last,the students’attention state is obtained through the feature fusion strategy,and students’attention time is calculated.The experimental results show that the algorithm has high accuracy and achieves the desired effect.

关 键 词：人脸检测 头部姿态估计 眼部疲劳 注意力检测 

分 类 号：G434[文化科学—教育学] TP18[文化科学—教育技术学] TP391.4[自动化与计算机技术—控制理论与控制工程]