Research On Intelligent Analysis Technology Of Teacher’s Behavior On Blackboard

Teaching quality is the lifeline of education development,and teaching quality supervision and evaluation is an effective means to ensure and improve teaching quality.In the education system of primary and secondary schools,teachers’ writing on the blackboard is an important part of teaching quality evaluation,which is related to teachers’ professional development and teaching quality evaluation criteria.The blackboard writing time of teachers and the quality of blackboard writing play an important role in the teaching effect,so they have become an important evaluation standard for the quality of teachers’ classroom teaching.In view of the current evaluation of teachers’ writing on the blackboard in primary and secondary schools,traditional methods such as supervisors watching teaching videos in class or after class for manual recording and analysis are still mainly used,which not only consumes a lot of labor time cost,but also is limited by supervisors.In addition,when there are multiple people writing on the blackboard in the real teaching scene,it is impossible to directly detect the changes of the blackboard to realize the classification and statistical analysis of the writing behavior of different people.Based on the above research background,this thesis focuses on the identification and analysis of a specific target(teacher)’s blackboard writing behavior in the real teaching scene where there are many teachers and students writing on the blackboard.Relevant research has been carried out,and the main research contents and research results are as follows:(1)In terms of specific target recognition and tracking algorithm research,Thanks to the excellent target detection performance of the YOLOv3 model,Siam RPN++ can track a specific dietary target in real time.To implement automatic detection and tracking of specific targets,we propose an automatic teacher detection and tracking algorithm based on the YOLOv3-Siam RPN++ model.The algorithm first uses YOLOv3 to detect specific targets based on the decision rules,and then automatically inputs the detected specific targets as search content into the Siam RPN++ model,thereby completing the real-time tracking of specific targets.Experimental results show that the accuracy of the model is significantly higher compared to the traditional algorithm,recall and real-time performance of specific target recognition and tracking,and it is more helpful to solve the inconsistencies of the traditional model.Adapt to problems such as long-term tracking.(2)In the research of gesture recognition algorithm for key parts of specific target,the complex gesture recognition problem is firstly transformed into a multi-classification problem of gestures.On this basis,an improved image classification algorithm based on Efficient Net technology is proposed for specific areas of interest.The model takes the detected region of the specific body part most directly related to the blackboard writing behavior as the region of interest,and uses this as the input image of the classification model.In the process of model improvement,first of all,The structure of the Efficient Net-B4 model was adapted and optimized,as it cannot successfully meet the real-time operational requirements due to the large number of parameters and high computational complexity.Then,the classification accuracy of the model is further improved by introducing a spatial attention mechanism.The test results show that compared to the original Efficient Net-B4 network,the improved Efficient Net-B4 model has obvious improvements in the accuracy of image classification,the training speed of the model,and the real-time performance of image classification in the region of interest.(3)In the research of teacher’s blackboard writing behavior recognition method,based on the research results obtained in this thesis in specific target recognition and tracking,specific target key part gesture recognition and tracking,an automatic recognition and analysis method of teacher’s blackboard writing behavior based on local feature matching is proposed.The method firstly constructs the behavior judgment rules and template library according to the statistical laws of the postures of the relevant parts of the body when the teacher writes on the blackboard.Then,using the output results of the teacher identification and tracking model,the specific region of interest extraction model,and the image classification model,its behavior pattern expression is automatically generated in real time according to the template.Finally,the template matching(search)algorithm is used to determine whether its behavior is blackboard writing behavior.Experimental results on the first prototype of the system show that the method proposed in this thesis has high recognition accuracy and speed.After further improvement,it can be applied to the intelligent recognition and analysis system of teachers’ blackboard writing behavior.And achieve the expected recognition effect.