Facial pose estimation is a very important requirement in analyzing the state of students in classrooms. To improve the accuracy of facial pose estimation in classroom scenarios, this paper proposes a network model for facial pose estimation based on Transformer, FPTR. The model directly estimates the 6DoF in an end-to-end manner and generates the facial bounding box without the need for face keypoint detection. The model is used for facial pose estimation of students in an online classroom and achieves test results that meet expectations. The application experiment of the actual classroom scene show that the method can effectively solve the problem of large face pose angle deviation predicted by the classical direct regression pose estimation method. To evaluate the effectiveness and performance of the proposed FPTR network model, multiple sets of analysis experiments including comparative analysis experiments and ablation analysis experiments are set up and analyzed. The experimental results show that the FPTR network model is capable of end-to-end facial pose estimation, directly outputs the 6DoF of a given input image without a pose adjustment step, and generates a facial bounding box.