Research On Intelligent Recognition Of Students’ Classroom Behavior Oriented To Video Streaming

The traditional classroom is still the main front of education and teaching,and the analysis of traditional classroom teaching is usually based on time sampling observation and teaching event analysis.However,the current quantitative analysis based on "time sampling",the analysis indicators were cumbersome and rigid,and the analysis standards tend to be complex,resulted in time-consuming,labor-intensive and low-efficiency research processes.The qualitative analysis based on "event analysis",the research was subjective,the professional level of the analyst was required.In addition,the research ability of large-scale analysis was weak,and there was a lack of continuous research.The current development direction of education and the focus of scholars were developing in the direction of intelligence,personalization,precision,and ubiquity.The realization of efficient and intelligent classroom teaching observation through artificial intelligence has also became an inevitable demand for the development of the era of education informatization.In the past,evaluators usually watched recorded and broadcast classroom videos after class to analyze students’ classroom behavior,which was time-consuming and inefficient.Therefore,evaluators keep up with the pace of education informatization and begin to try to use intelligent technical means to carry out evaluation and analysis of teaching classroom behavior.However,in the field of intelligent classroom behavior recognition,there were still no standard coding methods and open large-scale datasets available,which directly lead to the difficulty of automatic identification of teaching and learning behaviors in traditional classrooms,and the low recognition accuracy.Many technical bottlenecks need to be solved.The difficulty of collecting student datasets,the complex background of the real classroom environment,the complex types of student behaviors and movements,and the students’ clothing and posture all increase the difficulty of intelligent recognition of students’ classroom behaviors to a certain extent.The advent of deep learning has compensated for the difficulty in implementing solutions for complex scenarios such as classroom behavior recognition.It can make full use of a large number of existing classroom teaching video resources,extract the human skeleton from the video for pose estimation training,and realize the classification and intelligent recognition of classroom behavior.To solve the above problems,this paper tried to extract the skeleton features of students by using human pose estimation and trained deep neural network to intelligently recognize students’ classroom behaviors.It also proposed a class behavior recognition method based on human skeleton and deep learning,which could help teachers to accurately grasp the students’ classroom learning.It was helpful for the development of intelligent teaching management and teaching evaluation.The specific research work was as follows:(1)This paper attempted to reduce the workload of teaching staff through deep learning and computer vision methods,and to intelligently identify students’ classroom behavior.The framework of selecting skeleton detection model Open Pose to extract human skeleton features to realize the recognition of students’ classroom behavior.(2)This paper constructed a multi-person classroom behavior dataset of students,with a total of 16,628 classroom behavior images,included six classroom behaviors:putting＿hand,listening,standing＿up,sleeping,looking around,and taking notes.(3)This research built a video stream-oriented intelligent recognition process and method for students’ classroom behaviors.This method could accurately identify the six classroom behaviors of students.The identification results were statistically analyzed,included student number,action name,action start time,action end time,action duration,etc(4)This study analyzed students’ learning engagement according to the statistical results,so that teachers could have sufficient time to reflect on their own shortcomings and problems in the teaching process according to students’ performance in class,and help teachers better grasp the whole class.