Calculation method of classroom head-up rate based on head pose estimation

The head-up rate is an important indicator to evaluate the quality of classroom teaching, which can reflect the students' class status in time. With the continuous development of computer vision, many techniques are gradually applied to classroom scenarios. By estimating the student's head posture, it can truly reflect the quality of classroom teaching and effectively reduce the delay of classroom feedback. However, the existing classroom environment is relatively complex, the faces captured by the surveillance cameras are slightly deformed, the sizes are different, and there are occlusion problems, which pose a great challenge to the estimation of the head pose. In response to these problems, this paper improves the RetinaFace model of face detection and introduces the Convolutional Block Attention Mechanism(CBAM) in the backbone network ResNet50 to improve the performance of the network. The Refinement Feature Pyramid module adopted in it can further solve the problem of inconsistent face size. Through the PNP-based method to estimate the head posture, calculate the class head-up rate, and objectively reflect the teaching quality. Experiments show that the method in this paper can achieve higher-precision head pose estimation on the public datasets WIDER FACE and FDDB.