Under the background of educational informatization, using information technology to analyze teaching behavior can promote the development of teaching activities and improve the efficiency of classroom evaluation. The distribution of students’ seats is a part of students’ classroom behavior, and it reflects the students’ interest in listening during the class. Thus, analyzing the distribution of students’ seats can reflect the attraction of the course. In view of the dense number of people, many small objects, and uneven illumination in classroom videos, and the large amount of redundancy of detection boxes, we propose an improved model suitable for the classroom environment to detect the upper bodies of students and seats by combining Weighted Boxes Fusion (WBF) with Yolo v5. Based on the prior conditions of camera position, seat arrangement, and classroom size, a new matching method, the point pair matching, for matching students with their seats and drawing the distribution chart of students’ seats, is proposed. We also carry out statistics on the distribution area of students’ seats. The improved Yolo v5 model is compared with Yolo v3, Yolo v4 and Yolo v5, and the proposed point pair matching method is compared with some traditional matching methods. The results show that the overall solution proposed in this paper has significant improvements in performance, and this method can solve the seat distribution problem of this type of seat with a separate backrest in seedling-style classrooms.