基于时空图卷积网络的课堂教学行为分析  

Analysis of Classroom Teaching Behavior based on Space-time Map Convolution Network

作　　者：高杨[1] GAO Yang(School of Public Security Information Technology and Intelligence,Criminal Investigation Police University of China,Shenyang,Liaoning,110034,China)

机构地区：[1]中国刑事警察学院公安信息技术与情报学院,辽宁沈阳110034

出　　处：《新疆师范大学学报（自然科学版）》2023年第1期89-96,共8页Journal of Xinjiang Normal University(Natural Sciences Edition)

基　　金：2021年度辽宁省普通高等教育本科教学改革研究一般项目(20211017502)。

摘　　要：文章通过研究基于时空图卷积网络的课堂教学行为,提升课堂教学行为分析效果。首先,通过区域生成网络,确定课堂环境内多个师生目标的区域;利用基于序列多尺度特征的多目标姿态估计方法,在确定的目标区域内,估计多个师生目标的姿态,获取全部师生的骨骼序列关节点位置信息;其次,引入时空图卷积网络,构建识别方法,在时空图卷积网络内输入骨骼序列关节点位置信息,输出课堂教学行为识别结果;最后,通过S-T分析法结合课堂教学行为识别结果,完成课堂教学行为分析。实验证明:该方法可有效确定多个师生目标区域,完成多目标姿态估计。在不同课堂教学行为时,该方法多目标姿态估计的精度较高,可有效分析课堂教学行为。Research classroom teaching behavior based on space-time map convolution networkto improve the effect of classroom teaching behavior analysis.Firstly through the regional generation network,determine the areas of multiple teacher and student goals in the classroom environment;The pose estimation method based on multi-scale features of the sequence is used to estimate the pose of multiple teachers and students in the determined target area,and obtain the joint point position information of all teachers and students'skeleton sequences;Secondly introduce spatiotemporal graph convolution network to construct recognition method,input the position information of joint points of bone sequence in the space-time map convolution network,and output the results of classroom teaching behavior recognition;Last through the S-T analysis method combined with the results of classroom teaching behavior identification,complete the analysis of classroom teaching behavior.The experiment shows that this method can effectively determine multiple target areas for teachers and students,and complete multi-target attitude estimation;In different classroom teaching behaviors,the accuracy of multi object attitude estimation is high.This method can effectively analyze classroom teaching behavior.

关 键 词：多目标 姿态估计 课堂教学 行为分析 多尺度特征 S-T分析法 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]