Video-based Student Behavior Analysis System

With the continuous improvement of the requirements for students 'personalized education,the focus of education and teaching gradually changes to the direction of students' personalization and specialization.In the teaching process,teachers cannot pay attention to the classroom behavior status of each student,they can only draw the quality of the student's classroom learning status from the students' homework or exams,and some subjective factors can easily make the teacher judge the student's learning status There is bias.In the traditional teaching process,the identification and control of student behavior is mainly carried out manually by the teacher,but this method not only consumes time and effort,but also affects the quality of the teacher 's class.How to realize the automatic recognition of student classroom actions has become a Problems to be solved.In view of the above problems,this article uses artificial intelligence technology and computer vision technology to identify the behavioral status of students in classroom scenes to help teachers obtain teaching feedback information.This article first discusses the existing bone key point detection methods,analyzes the selected Open Pose model in detail,and expounds the bone key point detection and connection process.Then,in response to the problem that there is no public data set for class action behavior recognition data,we collect teaching videos in classroom scenes and construct standard action data sets,which are divided into five categories: raising hands,lying down,supporting cheeks,playing mobile phones and writing.After performing key point analysis on these five types of actions,the weakly related bone key points are removed and key point extraction and normalization are performed through the Open Pose model.By comparing the two classification models of random forest and support vector machine on the data set The classification effect of the final selection of polynomial kernel support vector machine is used as the method of model training and classification in the scene where the key points of the bone are clear.At the same time,this paper compares the action recognition method based on deep learning migration with the above recognition methods,and proposes a deep learning migration action recognition method based on the grayscale heat map of bone key points to solve the action of the situation where the key information of the bone is incomplete Identify the problem.The fusion of this method with the skeleton key point action recognition method can greatly enhance the effectiveness of action recognition.Finally,based on the analysis of the above research content and recognition methods,the video-based student behavior analysis and demonstration system is designed and implemented on the Qt development platform.