Research And Implementation Of Student’s Behavior Recognition Algorithm In Classroom

With the development of information technology,the modes of teaching have also been changed.In addition to traditional methods of teaching,many new types such as online teaching and smart classrooms are becoming popular.In the era of intelligent education,if information technology can be used to implement a class analysis system that could automatically analyze behaviors of students in the class and provide teachers with results,it will give lots of helps to improve teaching effectiveness and boost educational equity.This paper designed a robust recognition algorithm for the students’ learning behaviors in real teaching scenarios of the primary school.It uses cameras to get the video data of classes in real time,automatically analyzes the class concentration information and the actions information of students,provides teachers with visualized and quantified results.Specifically,this paper makes contributions in the following four aspects:(1)For the task of recognizing the class concentration,we established a student detection dataset and designed a student detection algorithm.The precisions of recognizing gazing and nongazing state are 0.91 and 0.92 and the recalls of recognizing gazing and non-gazing state are 0.90 and 0.93.This algorithm achieves the best performance at present.(2)For the task of recognizing the students’ actions,we established a students’ actions dataset and designed an action recognition algorithm.The precisions of three actions are 0.878,0.890 and 0.887,and the recalls are 0.753,0.831 and 0.850.This algorithm achieves superior accuracy over the state-of-the-art techniques.(3)For the task of processing the learning behaviors information of students,we designed a reasonable quantification and visualization strategy.We proposed a gaze rate calculation method to quantify the class concentration information and the method for judging the status of the class discussion.The accuracy of the algorithm for counting the number of students and judging the status of the class discussion are 0.991 and 0.993.This algorithm can meet the accuracy requirements in real scenarios.The experimental results show that the algorithm achieves the best performance in current algorithms on the task of recognizing the students’ learning behaviors in real teaching scenarios of the primary school.Our algorithm can provide data support for teachers and researchers to carry out teaching research.