The Research And Implementation Of A Classroom Attention Model Based On Computer Vision

Benefitting from the rapid development of information technology,“smart learning”and “smart education” have become an important part of educational development.Smart classroom,which is based on big data and artificial intelligence technology,has been building in many universities.How we understand “smart” classroom is a valuable topic which attaches many researchers’ attention.As a key point of smart classroom research,a scientific,objective,and intelligent learning evaluation is an essential part to represent“smart”.This thesis uses computer vision technology to study classroom attention.First,this thesis proposes a set of classroom attention modeling methods based on head posture detection.Then it implements the detection algorithm considering the specific problems which lecture-based classroom and seminar-based classroom face.Finally,it carries out verification on real classroom scene.The main contributions of this thesis include:(a)This thesis proposes a very complete classroom attention measurement method,which consists of two parts: the head pose detection algorithm and the attention model.The head pose detection algorithm provides data support for the whole program.The attention model,which is used to quantify classroom attention,is composed of head-up rate,activity coefficient and attention-K value.As an attention parameter proposed by this thesis,the activity coefficient can describe the classroom attention in great details.(b)To addressing the inaccuracy of head posture detection in the back rows of lecture-based classroom,a two-stage head posture detection algorithm is proposed.The precision of the student number detection is 0.978,while the recall is 0.984,and the precisions of head-up and head-down detection are 0.92 and 0.91,while the recalls are 0.90 and 0.94.This algorithm has achieved the state-of-the-art on lecture-based classroom detection.(c)To addressing the problem that the head posture detection algorithm of lecture-based classroom cannot work in the seminar-based classroom,a specific head posture detection method is proposed.The precision of the student number detection is 0.96,while the recall is 0.96,and the precisions of head-up and head-down detection are 0.88 and 0.86,while the recalls are 0.87 and 0.87.This thesis applies the attention model to analyze a compulsory course of the College of Computer Science and Technology in 2019.After analyzing and discussing with teachers,the results of the model visualization and the attention-K value are consistent with subjective feelings of teachers,which illustrates that the attention model has application value in describing the classroom attention situation.