《计算机应用研究》|Application Research of Computers

基于改进OpenPose的学生行为识别研究

Research on student behavior recognition based on improved OpenPose

免费全文下载 (已被下载 次)  
获取PDF全文
作者 苏超,王国中
机构 上海工程技术大学 电子电气工程学院,上海 201600
统计 摘要被查看 次,已被下载
文章编号 1001-3695(2021)10-054-3183-06
DOI 10.19734/j.issn.1001-3695.2020.11.0435
摘要 学习者的行为动作能够反映出学习者的学习状态。传统学习者行为识别方法存在实时性不高、准确率低等问题。针对以上问题,提出了一种基于改进OpenPose的学习者行为识别方法。该方法从人体骨骼关节点角度出发,首先通过高斯滤波消除图像中的高斯噪声;然后通过融入注意力机制的目标检测算法检测图像中的目标学生位置,再将检测后的图像通过改进的OpenPose模型提取人体骨骼关节点坐标;最后利用ST-SVM分类器对获取的关节点坐标进行分类,从而快速准确地识别出学习者的行为状态。实验结果表明,该方法能够快速、准确地识别出学生的行为动作,准确率达到99%以上,fps达到了20以上,相比原模型,fps提升了50%。
关键词 行为识别; 骨骼关节点; Tiny_YOLOv3; OpenPose; ST-SVM
基金项目 国家重点研发计划资助项目(2019YFB1802700)
上海工程技术大学研究生创新计划资助项目(19KY0232)
本文URL http://www.arocmag.com/article/01-2021-10-054.html
英文标题 Research on student behavior recognition based on improved OpenPose
作者英文名 Su Chao, Wang Guozhong
机构英文名 School of Electronic & Electrical Engineering,Shanghai University of Engineering & Science,Shanghai 201600,China
英文摘要 The learner's behavior can reflect the learner's learning state to a certain extent. Traditional learner behavior re-cognition methods have problems such as low real-time performance and low accuracy. Aiming at the above problems, this paper proposed a learner behavior recognition method based on improved OpenPose. This method was considered from the point of view of human bone joints. Firstly, it removed the Gaussian noise in the image through Gaussian filtering, and then detected the target student position in the image through the target detection algorithm integrated into the attention mechanism, and then extracted human bone joint point coordinates from the detected image through the improved OpenPose model, and finally used the ST-SVM to classify the acquired joint point coordinates to quickly and accurately identify the learner's behavior state. The experimental results show that the method can quickly and accurately recognize the student's actions, the accuracy rate is over 99%, and the fps is over 20. Compared with the original model, the fps is improved by 50%.
英文关键词 behavior recognition; bone joint points; Tiny_YOLOv3; OpenPose; ST-SVM
参考文献 查看稿件参考文献
 
收稿日期 2020/11/20
修回日期 2021/1/4
页码 3183-3188
中图分类号 TP183
文献标志码 A