Research On Classroom Behavior Recognition Of Students Based On Human Skeleton And Deep Learning

The classroom behavior of students is an important part of classroom teaching evaluation,which is of great significance to the improvement of teaching quality.In traditional classroom teaching,the teacher only observes students' classroom behavior in class,and after class,students' classroom behavior is analyzed by watching classroom videos or analyzing classroom teaching videos based on coding system.These methods rely on observers to analyze students' classroom behavior,which is time-consuming and inefficient,so it is impossible to conduct large-scale and continuous observation and measurement.In recent years,with the rapid development and large-scale application of artificial intelligence technology and computer vision technology and the deepening of educational reform,more and more attention has been paid to the information and intelligent analysis of students' classroom behavior.The behavior of the traditional recognition method using the artificial design characteristics and training classifier is used to identify the behavior,only applies to the scenario simple,single,less action classification such as small target classification problem,and the recognition rate are greatly influenced by the characteristics of the artificial design,not suitable for the environment is complex,multiple objective behavior recognition,unable to meet the needs of students' classroom behavior recognition.Behavior recognition method based on the deep learning while overcoming the shortcomings of traditional method,but because there is no open large-scale student classroom behavior data set,in the student classroom behavior under the condition of limited data sets,use only the depth of learning in a way that end-to-end,susceptible to interference factors of the real environment,such as the body posture,gender,subtle movements of degree,the complexity of the actions and similarity and so on,so it is hard to satisfy the classification of all students' classroom behavior.Aiming at the above problems,this paper uses pose estimation to extract the characteristics of human skeleton and train the deep neural network to identify students'classroom behavior,and proposes a classroom behavior recognition method based on human skeleton and deep learning,which can timely and effectively reflect students'learning status,help teachers accurately grasp students' classroom learning,help intelligent teaching and management,and improve the quality of classroom teaching.The specific research work is as follows:(1)This paper studies the relevant theories and techniques of human pose estimation technology to obtain human skeleton,introduces two typical and common skeleton detection models AlphaPose and OpenPose,and makes a comparative analysis of the two models,and selects the OpenPose skeleton detection model to extract human skeleton for students' classroom behavior recognition.(2)Since there is no publicly available data set of students' classroom behavior,this paper uses video camera to collect 4,800 classroom behavior images of students in an experimental primary school in Xiamen,including 7 typical classroom behaviors such as raising hands,attending classes,looking around,reading,writing,standing up and sleeping,which constitute Student Classroom Behavior Image Database,hereinafter referred to as SCBID.(3)To rule out real environment factors,and further improve students' classroom behavior recognition accuracy,this paper studies the students classroom behavior recognition method based on the human body skeleton and deep learning,by extracting the student classroom behavior image of human body skeleton,and combined with this paper established a 10 layer depth of the convolutional neural network CNN-10 to students'classroom behavior identification.(4)Cnn-10 network and the method proposed in this paper were used for training and testing,and the accuracy was 93.65%and 97.92%.The results show that the method in this paper is better than cnn-10 based on deep learning.At the same time,AlexNet,VGG16,GoogleNet and ResNetl8 were used in SCBID database to identify students' classroom behaviors,and the recognition effect was compared with some representative classroom behavior recognition methods,proving that this method can effectively identify students'classroom behaviors and improve the recognition rate.