计算机工程与应用 ›› 2022, Vol. 58 ›› Issue (6): 164-169.DOI: 10.3778/j.issn.1002-8331.2009-0510

• 模式识别与人工智能 • 上一篇    下一篇

课堂环境下用于头部行为识别的李群特征表示

谢冬,孟凡荣,贺恒桃,闫秋艳   

  1. 中国矿业大学 计算机科学与技术学院,江苏 徐州 221116
  • 出版日期:2022-03-15 发布日期:2022-03-15

Lie Group Feature Representation Method Applied to Head Behavior Recognition in Classroom Environment

XIE Dong, MENG Fanrong, HE Hengtao, YAN Qiuyan   

  1. College of Computer Science, China University of Mining and Technology, Xuzhou, Jiangsu 221116, China
  • Online:2022-03-15 Published:2022-03-15

摘要: 头部行为是个体行为的重要组成部分,在课堂环境下对于学生的行为来说更是如此。使用传统的RGB视频图像进行头部行为识别有着许多限制,例如背景的干扰和光线的变化等,而深度图像可以通过包含的深度信息很好地处理这些问题。针对课堂环境下的头部行为识别问题,受到李群理论的启发,提出了一种从深度图像中提取李群特征表示的模型,并且使用该李群特征完成了头部行为识别任务。从深度图像中获取脸部的关键点及关键段信息,通过计算相邻帧之间关键段的旋转及位移获得能够同时表示时间空间信息的李群特征表示,使用支持向量机来完成头部行为的分类识别。在公开数据集上验证了方法的有效性,然后通过Kinect获取制作了课堂环境下的真实行为数据,实验结果表明李群特征表示方法能够有效帮助课堂环境下头部行为的识别,对课堂环境下的学生行为识别提供了帮助。

关键词: 头部行为识别, 李群特征, 课堂环境

Abstract: Head behavior is an important part of individual behavior, especially for students’ behavior in the classroom environment. There are some well-known limitations in using RGB images for head behavior recognition such as background clutter and illumination changes. Depth maps can handle these problems well with the included depth information. Aiming at the problem of head behavior recognition in the classroom environment, inspired by Lie group theory, this paper proposes a new feature representation model based on Lie group from depth map, and designs a head behavior recognition method based on this feature representation. Key points and segments of face are constructed from 3D coordinates in depth map. The Lie group feature is extracted from the relationship between the key segments of adjacent frames, which can represent the spatial and temporal relationship of head behavior at the same time. Support vector machine(SVM) is applied to identify head behavior in this method. This work verifies the effectiveness of the proposed method on public data sets. Then the real behavior data in the classroom environment is obtained through Kinect. The experimental results show that the Lie group feature representation method can effectively help the recognition of head behavior in the classroom environment. The method can provide help for the recognition of student behavior in the classroom environment.

Key words: head behavior recognition, Lie group feature, classroom environment