Research On Head Pose Estimation Method For Learning Behavior Analysis In Smart Classroom

With the continuous integration of the "Internet Plus" education and the intelligent technology,the multi-screen and multi-touch teaching mode under the scenario of intelligent classroom arises at the right moment.The teaching content and the expression form of teaching activities will be more colorful.However,the teacher-centered cramming teaching model is not suitable for the teacher-student interaction in the new era,which is not conducive to the improvement of teaching effect.In order to guide students to become the main body of teaching activities.it is necessary to master the learning state and interest of students in the teaching process.It is the key to improve the teaching quality that students attention in class can objectively and truly reflect students interest,knowledge blind spot and positive state in teaching activities.As the main embodiment of students attention direction,this thesis will carry out research on the estimation method of head pose according to the scenario of wisdom classroom.Head pose estimation is a hot research topic in the field of computer vision,which has a wide range of application values and prospects.In recent years,more and more researchers have paid attention to it.In this thesis,a new head pose database with higher applicability is established according to the unique characteristics of intelligent classroom scenes and the existing problems of the head pose database.On this basis,the advantages of deep learning are utilized to improve the performance of the head pose algorithm,construct the intelligent classroom attention discrimination method for students,and effectively prompt and intervene in teaching activities,so as to improve the quality of teaching.The main work of this thesis is embodied in the following aspects:(1)A non-uniform head pose database(IRHP database)for intelligent classroom is developed to address the problem that the existing head pose database cannot meet the needs of intelligent classroom application.Firstly,two data acquisition methods of 15°interval and 5° interval are designed in the aspect of data acquisition scene in this thesis.Secondly,the selection of infrared camera equipment in the acquisition equipment can effectively avoid the impact of light changes.Finally.in the data label,the whole process of using multi-person assistance,manual production of labels,to ensure the authenticity and effectiveness of the database label.(2)A series of multi-feature analysis head pose estimation baseline methods based on the IRHP database are implemented to prove the applicability and application value of the IRHP database proposed in this thesis.On the one hand,the traditional machine vision and image processing technology is used to extract the face image features of a variety of filter operators,compare and analyze their advantages and disadvantages,construct the appropriate feature vectors,and input them into the machine learning classification method to predict the head pose angle.On the other hand,the convolutional neural network model is built,and the large-scale face image database is used for pre-training of the model,and the effective face key features are obtained to extract spatial parameters.Then,the experimental fine-tuning is carried out based on the IRHP database to improve the generalization ability and robustness of the model.(3)A head pose estimation method based on multi-scale feature information fusion(IRHP-Net)is proposed.Firstly,according to the characteristics of the task in this thesis,the head pose estimation task is mathematically expressed,the appropriate loss function is set,and the regularization technology is used to improve the generalization ability of the model.Facing the convolutional neural network features information loss in the process of convergence problem,analyze and compare different scales in the process of convolution,explore and put forward the characteristics of three kinds of fusion operator(the CW,WF and CVF),the high-level image semantics information and low-level abstraction features,the combination of trying to make up for missing feature information in the process of training.In the training process,batch gradient descent algorithm is introduced to optimize the model parameters,and the experimental results demonstrate that this model has better performance than other methods.(4)Based on the above head pose estimation algorithm,the attention discrimination scheme of students in the intelligent classroom learning environment is designed.Through head pose estimation algorithm to identify the students personal attention area,using the camera coordinate system and world coordinate system to establish the geometric model,will gather in obtaining personal attention area of joint attention area,distraction index in construction and setting the threshold parameter,the distraction index more than threshold student criterion in the state of concentration,for teachers to make relevant teaching measures.