Research On Classroom Human Behavior Recognition Based On Convolutional Neural Network

Classroom education is an important part of school education.How to improve the quality of classroom education has always been an important research content of school education.In recent years,with the rapid development of computer technology and computer vision technology,classroom intelligence has also become the focus of classroom education development.The study of classroom human behavior recognition is a key field of classroom intelligence.The behavior of students in the classroom reflects the degree of teacher-student interaction,the enthusiasm of students in class and the ability of teachers to teach.Understanding and analyzing the students’ behaviors in the classroom is helpful to understand the status of students during class,so as to improve the teaching methods and improve the quality of teaching.Therefore,the study of human behavior recognition in the classroom is a meaningful and challenging subject.Although there has been more in-depth research on human behavior recognition,the research on human behavior recognition in the classroom scene is still in the development stage.This paper studies the behaviors of students in the classroom based on convolutional neural networks such as standing up and raising their hands.The main work of the thesis is as follows:(1)Constructed a data set for classroom standing.The original data of the data set comes from network resources and real classroom recordings.The data set is divided into three large-scale,small-scale,and small-scale classroom scenes according to the number of students in the classroom.Each type of classroom contains different light source conditions.The constructed data set is used for training and testing when students stand up.(2)Student identity verification based on face recognition.The convolutional neural network is applied to face recognition in the classroom.For the similar characteristics of faces,the loss function of VGG-16(Visual Geometry Group)is improved to reduce the intra-class distance and increase the inter-class distance.The improved VGG-16 network improves the accuracy of face recognition.The accuracy rate on the Labeled Faces in the Wild(LFW)data set is 2.4% higher than that of the classic VGG-16.It can effectively verify the identity of classroom students who have experienced standing up.(3)Recognition of raising hands in classroom based on key points of bones.Aiming at the diversity of the hands raised by the students in the classroom,the method of detecting the key points of the human bones is adopted to judge the behavior of the students raising the hand high in the classroom.First,the Open Pose model is used to detect the key points of the skeleton of the students in the classroom.Using the coordinate information and connection information of the detected key points,the posture of the students’ arms is analyzed,and finally the purpose of recognizing the behavior of the students raising their hands is reached.(4)Classroom standing behavior recognition based on improved SSD(Single Shot Multi Box Detector).Since the students in the classroom are in a sitting posture during class,the students ’standing behavior is determined by detecting the students’ standing state.The basic network structure of the SSD algorithm is improved,the number of network layers is deepened,and the network feature extraction capability is enhanced.Aiming at the problem of unbalanced positive and negative samples in the classroom,the loss function is improved.The improved SSD algorithm has an accuracy rate of 86.7% on the self-built data set and a detection speed of 33fps(Frames Per Second),which meets the real-time requirements.