System Of Classroom Student Behavior Analysis Based On Computer Vision

In October 2020,China released a document named Overall Plan for Deepening the Reform of Educational Evaluation in the New Era,which states that education evaluation should focus on the process,abandon the results-only theory,and make the education evaluation more diversified and diverse.Nowadays,teachers in China mainly teach in classes,which is a one-to-many teaching style,and it is difficult for teachers to evaluate each student’s performance in depth and comprehensively in the classroom.With the development of computer vision,behaviour recognition methods based on surveillance video are used in various fields,but the development in the field of educa-tion is slow.Using of computer vision technology to analyse student behaviour in the classroom not only reduces the pressure on teachers,but also makes the educational eval-uation of classroom sessions more accurate and scientific.The video of classroom scenes is characterised by dense crowding,severe occlusion and serious deformation,so the fea-ture extraction of the human body in classroom videos through neural networks often has a large error in human behaviour recognition.This paper focuses on a computer vision-based method for classroom student be-haviour analysis.It can be summarized.(1)This paper uses the key points of the human body as the basis for behavioural recognition.According to the magnitude of body skeleton movement,the classroom stu-dent behaviour is divided into large-amplitude action behaviour and small-amplitude ac-tion behaviour,and different recognition methods are used for different behaviours.(2)This paper proposes a large-amplitude behaviour recognition method based on human key points.The method first detects all the human key points in the image and uses the human key points as the basis for behaviour recognition.Then,the support vector machine method is used to classify the coordinates of the human key points.In this paper,four types of behaviour based on this recognition method are carried out: head down,side body,hand up and listening,with good behaviour recognition results.(3)A small-amplitude behaviour recognition method based on object detection is proposed.The method detects four objects: hand,mobile phone,laptop and book by YOLOv3,calculates the intersection ratio between the hand and the other three objects to measure the association between the hand and each object,and detects each behaviour according to the association ratio.(4)A Java web-based student behaviour analysis system is developed.The system allows users to upload classroom videos or classroom picture sequences,and automat-ically records students’ classroom behavioural performance through the processing of a behavioural recognition model,which provides data management,data correction and data analysis functions.(5)This paper establishes a high-quality dataset for student behaviour recognition.The dataset is derived from high-definition camera equipment,and the dataset are images,each of which contains only one person in the classroom,containing two parts of annotated information: coordinates of human key points and behavioural category labels? target de-tection frames and corresponding categories of people and objects in the classroom scene.In summary,this paper provides an automated system for processing and analysing student behaviour in the classroom based on human keypoint detection technology and target detection technology,which performs well in terms of accuracy and timeliness and provides a convenient tool for classroom education evaluation.