Research On Learning Analysis Based On Classroom Video

Learning analysis is a technique that counts the data generated during the whole process of students' learning and discovers the rules to improve the learning effect.With the advent of the era of big data,data mining,deep learning and other technologies continue to develop,providing technical support for learning analysis.In traditional teaching,students' classroom performance is mainly based on manual observation records based on the classroom observation scale.This method is inefficient,and manual observation is prone to errors.To solve this problem,we adopt the method of recording classroom videos to realize real-time recording of students' behavior in the classroom,through the analysis of classroom videos,we can fully grasp the students' class performance.Our work has done the following four parts of research around classroom video analysis.A learning analysis model based on classroom video was proposed.We summarized several common learning analysis models and found that learning analysis models are inseparable from three parts: data collection,data analysis,and teaching application.Among them,data collection is the basic part of learning analysis,and data analysis is the core technology of learning analysis.Application is the manifestation of the value of learning analysis.On these basis,according to the actual situation of this research,our work proposed a learning analysis model based on classroom video.The classroom behavior recognition method based on pose estimation was studied.In this paper,two methods,bottom-up and top-down,in pose estimation were respectively selected for research.Among them,the bottom-up model had missing key point detection and disordered body connections;the top-down model had higher accuracy than the bottom-up model,while the recognition time would increase with the increase of the number of people.Therefore,pose estimation is not suitable for the practical application of our research.A classroom video analysis model based on object detection and face recognition was designed.First,several face detection algorithms and face recognition algorithms were compared for classroom videos.Light and Fast Face Detector(LFFD)was selected as the face detection algorithm,at the same time,Arc Face was selected as the face recognition algorithm.Then we made small improvements to the YOLOv5 object detection algorithm,and the improved network was used as an action recognition network.It was applied to the analysis of the "Discrete Mathematics" classroom video.Finally,the accuracy of the model was verified with the observation scale of classroom behavior.A correlation analysis of students' performance and classroom behavior method based on the LASSO-LARS algorithm was designed.The LASSO-LARS regression method was used to analyze the correlation between students' classroom behavior and test scores,and then determined a student's classroom behavior scoring model.Finally,the intervention objects were determined according to the behavior scoring model.The learning intervention was performed on them,and achieved certain effects.