一种头部姿态估计的课堂注意力参数分析方法  

The Parameter Analysis Method of Classroom Attention for Head Pose Estimation

作　　者：王慧敏 刘洪甜 张建伟 WANG Huimin;LIU Hongtian;ZHANG Jianwei(Army Academy of Armored Forces,Beijing 100072,China)

机构地区：[1]陆军装甲兵学院,北京100072

出　　处：《信息工程大学学报》2025年第2期168-174,共7页Journal of Information Engineering University

基　　金：军事院校军事教育科研课题(2023QN04)。

摘　　要：针对研讨型或纵深较大的讲授型教室中头部姿态估计方法存在的误检、漏检问题,提出一种课堂注意力测量方案。首先,采用级联神经网络结构对课堂视频进行头部检测,精确获取学生头部的位置和数量。其次,构建全局人脸检测器和头部姿态分类器,结合人脸检测与姿态分类,准确判定凝视与非凝视状态。接着,引入活跃参数,通过前后帧头部位置匹配算法计算头部姿态变化次数,并结合凝视状态改变次数表征抬头率,得到新的抬头率计算公式。最后,结合抬头率和活跃参数,通过注意力模型分析得出总体表征课堂注意力的数值——注意力K值。实验结果显示,头部检测器在多尺度课堂中的准确率超95.6%,头部姿态分类器分类精确率达94.6%。实际课堂验证表明,该方案能真实反映学生注意力情况,为教学反思和教学能力评价提供重要参考。A classroom attention measurement scheme is proposed to address false and missed detections in head pose estimation within seminar or large-depth lecture classrooms.Firstly,a cascaded neural network is employed to detect heads in classroom videos in order to accurately locate and count student heads.Secondly,a global face detector and head pose classifier are developed to precisely determine gaze and non-gaze states by combining face detection and pose classification.Subsequently by introducing active parameters are introduced,and a head position matching algorithm is used to calculate head pose changes between frames.A new formula for head-up rate is derived based on gaze state changes.Finally,an attention model is analyzed to obtain an overall numerical value,the attention Kvalue,which combines head-up rate and active parameters.Experimental results indicate that the accuracy of the head detedor exceeds 95.6%in multi-scale classrooms,and the precision of the head classifier reaches 94.6%.Verification in real classrooms shows that the scheme accurately reflects student attention,providing valuable insights for teaching reflection and evaluation.

关 键 词：计算机视觉 课堂视频 头部姿态估计 注意力模型 抬头率 活跃参数 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]