Research on Head Detection and State Estimation Algorithm in Classroom Scene

Yuting Huang, Fan Bai, Chongwen Wang

Abstract: The penetration rate of mobile phones and tablet computers among college students is increasing, and the loose teaching environment has led to a large number of phubbers in college classrooms. The state of students' attendance in class is an intuitive indicator of classroom quality. Obtaining this data in real-time will bring great help to school evaluation and improvement of teaching standards. The data in this article comes from teaching videos collected by high-definition cameras in colleges. Through offline training, the face detector HDN can accurately extract the position coordinates of the student in the picture in the real teaching scene and pass the detected head information to the convolutional network responsible for judging the state of the student's head to obtain the student's current Class status. The HDN designed in this paper achieves a recall rate of more than 95% on the authoritative public dataset FDDB, and the accuracy of Wider Face's face dataset under three difficulty conditions is 93.9%, 93.2%, and 88.0%. The self-designed Raised Head Network achieves 88% accuracy on the RaisedHead dataset.