人体行为识别及在教育录播系统中的应用

发布时间：2018-06-19 12:06

本文选题：教育录播系统 + 人体行为识别　；参考：《西安科技大学》2017年硕士论文

【摘要】：基于视频图像技术的人体行为识别算法研究已经成为当前研究的热门课题,近些年取得了相当多的研究成果,其被应用到视频监控、道路交通、虚拟现实、医疗监护、体育运动等领域。另外,随着我国对教育事业的大力支持和高度重视,多媒体教学已经是广泛流行,其中对教学过程的录制是主要的发展方向,因此基于视频图像跟踪识别分析技术的智能教育录播系统应运而生。其改变了以往老师在讲台上讲课,学生坐在下面听课的教学模式,使得课后学生自己或者其他的学生可以随时学到知识,达到教学资源的共享和再利用。基于视频的教学录播系统主要包括教师跟踪系统和学生定位系统,本文主要是研究学生模块的学生课堂行为识别算法,以达到教育录播系统的实际需求。本文主要研究了传统的人体行为识别方法,以及对教室环境下行为识别算法的分析,对学生课堂上发言或者回答问题这一过程的行为进行识别,主要识别举手、站立和坐下三种动作。针对研究场景的背景以及学生动作本身的特点,本文提出一种基于运动历史图的Zernike矩特征和朴素贝叶斯分类器分类,然后通过Lucas-Kanade光流特征和全局运动方向特征判断运动方向的方法对动作进行识别。在前景提取方面,研究了几种常用的运动目标检测方法以及对比实验结果,根据本文研究的应用场景,使用背景减除法对前景进行检测;在特征提取方面,主要在运动历史图的基础上提取动作的Zernike矩特征,以及提取光流特征和全局运动方向等方向特征;在分类识别方面,主要使用朴素贝叶斯分类器对学生的三种动作分类识别,将举手动作和站立、坐下两种动作分为两大类;基于本文所拍摄的Student-behavior视频库进行实验测试,实验结果表明本文所提出的方法对背景复杂的教室环境下的学生行为进行有效识别,以及能够对模拟教师和学生干扰的场景准确识别,并且识别率高,具有一定的可行性和鲁棒性。
[Abstract]:The research of human behavior recognition algorithm based on video image technology has become a hot topic. In recent years, a lot of research results have been obtained. It has been applied to video surveillance, road traffic, virtual reality, medical monitoring, etc. Sports, etc. In addition, with the great support and high attention paid to education in our country, multimedia teaching has become a widespread trend, in which the recording of the teaching process is the main development direction. Therefore, the intelligent education recording and broadcasting system based on video image tracking recognition and analysis technology emerges as the times require. It has changed the teaching mode in which the teacher lectured on the podium and the students sat down to listen to the class, so that the students themselves or other students can learn knowledge at any time after class, so as to achieve the sharing and reuse of teaching resources. The teaching recording and broadcasting system based on video mainly includes the teacher tracking system and the student orientation system. This paper mainly studies the students' classroom behavior identification algorithm in order to meet the actual demand of the educational recording and broadcasting system. This paper mainly studies the traditional human behavior recognition method, and the analysis of the behavior recognition algorithm in the classroom environment, the identification of the students' behavior in the process of speaking or answering questions in the classroom, mainly the identification of raising hands. Stand and sit down. According to the background of the scene and the characteristics of students' actions, this paper presents a Zernike moment feature and naive Bayesian classifier based on motion history graph. Then the motion is recognized by Lucas-Kanade optical flow feature and global motion direction feature. In the aspect of foreground extraction, several commonly used moving target detection methods and comparative experimental results are studied. According to the application scene of this paper, background subtraction method is used to detect foreground. On the basis of motion history map, Zernike moment feature of action, optical flow feature and global motion direction feature are extracted. In classification and recognition, naive Bayes classifier is mainly used to classify students' three kinds of actions. According to the Student-behavior video library taken in this paper, the experimental results show that the method proposed in this paper can effectively recognize the students' behavior in the complicated classroom environment. And it can accurately identify the scene of simulated teacher and student interference, and the recognition rate is high, which is feasible and robust.
【学位授予单位】：西安科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：G434;TP391.41

【参考文献】