Comparison and Analysis of Student Behavior Recognition Methods in University Classrooms

Abstract

Student behavior intelligence recognition is a very important component of intelligent evaluation in university classrooms. Firstly, 1000 images containing various types of student behaviors in university classrooms were collected to construct a student behavior recognition dataset. These student behaviors include listen, read, write, bow, lie, yawn, drink, play, glance and trick. Then, classic CNN object recognition methods including SSD, Faster RCNN, YOLOV8, YOLOV11 and YOLOV12 were used to recognize student behaviors in university classrooms on the above dataset and the recognition performance of different methods was compared and analyzed. The experimental results show that YOLOV12 has the best student behavior recognition performance, with an mAP value of 0.521 and YOLO V11 has the fastest inference speed and the second highest recognition performance with an mAP value of 0.519.