基于人体骨架和深度学习的教师体态语言智能测评  

Research on Intelligent Assessment of Teachers’Body Language Based on Human Skeleton and Deep Learning

作　　者：王永固[1] 马家荣 王瑞琳 WANG Yonggu;MA Jiarong;WANG Ruilin(College of Educational Science and Technology,Zhejiang University of Technology,Hangzhou 310023,China)

机构地区：[1]浙江工业大学教育科学与技术学院,浙江杭州310023

出　　处：《开放教育研究》2023年第1期100-111,共12页Open Education Research

基　　金：2021年国家自然科学基金面上项目“多模态特征融合的自闭症教育机器人情感社交智能感知模型及应用研究”(62177043);浙江省2020年哲学社会科学规划重大项目“‘互联网+’残疾人社会支持智慧服务构建与创新”(20XXJC01ZD)。

摘　　要：深度学习技术与教师教育融合创新,可赋能教师教学体态语言行为精准测评。针对教师体态语言人工视频分析技术滞后问题,本研究采用人物目标检测、目标追踪和体态骨架图生成三种方法,创建真实教学情境下基于人体骨架图的专家型教师标准体态语言数据集,使用金字塔卷积模块优化ResNet50残差网络,构建金字塔残差神经网络模型,并基于数据集开展识别模型的训练、验证和测试实验。结果显示,体态语言识别模型的准确率、精确率和召回率均达到95%以上,识别准确率由高到低依次为工具性体态语言、巡视性体态语言、指示性体态语言、常规性体态语言、描述性体态语言。本研究基于上述训练良好的模型,开发教师体态语言智能感知测评系统,提出教师体态语言特征的人体骨架表征方法、教师体态语言数据集创建方法及智能识别神经网络模型构建方法,构建了大数据驱动的教师课堂体态语言智能测评方法体系。The integration and innovation of deep learning technology and teacher education enable teachers to accurately evaluate body language behavior in teaching.To address the problem of the lag of artificial video analysis technology of teachers'body language,this study used three methods of human-target detection,human-target tracking,and body skeleton map generation to construct the expert teacher's standard body language dataset based on the human skeleton map in the real teaching situation.The study also built a pyramid residual network model,which used the pyramid convolution module to optimize the ResNet50 residual network.Then,the study trained,validated,and tested the recognition model based on the dataset.The results showed that the accuracy,precision,and recall rates of the body language recognition model were more than 95%,and the recognition accuracy rates from high to low are instrumental body language,patrolling body language,indicative body language,routine body language,and descriptive body language.Based on the above-mentioned well-trained model,this study developed an intelligent perception assessment system for teachers'body language,and proposed the human skeleton representation method for teachers'body language features,a dataset creation method for teachers'body language,an intelligent recognition neural network model construction method,and the constructs of a big data-driven intelligent assessment system for teachers'body language in the classroom.

关 键 词：教师体态语言 智能识别 深度学习 人体骨架 神经网络模型 

分 类 号：G420[文化科学—课程与教学论]