Key points are not available for this paper at this time.
Classroom behavior analysis is an effective way to evaluate the teaching effectiveness in the field of learning analytics. However, traditional classroom behavior analysis mainly focuses on the teacher's observation or manual analysis of the classroom videos, which are time-consuming, laborious and subjective. In this paper, we design and implement a classroom students convergent behavior analysis system which based on image recognition. To adapt to the teaching scene, a student face detection method based on MTCNN (Multi-task Cascaded Convolutional Networks) and a student head pose estimation method based on SSR-Net are proposed respectively. The face detection method is improved through NMS (Non-Maximum Suppression), pooling and convolution to alleviate the problems of partial occlusion, variable posture, small scale and large number of students in the classroom environment. For head pose estimation, we embed the ECA (Efficient Channel Attention) mechanism to improve detection accuracy and speed. We use the face detection and head pose estimation methods to identify the behavior of the student's head-up and then analyze the convergence of students. In the experiments, we first demonstrate the head-up detection approach which is the basic of the convergent behavior analysis is feasible and strong timeliness. Then, the equal interval sampling experiments of different classrooms prove that the convergence behavior analysis of the head-up can accurately feedback students' classroom learning and is practicality for teaching evaluation.
Wang et al. (Mon,) studied this question.