Research on video capture scheme and face recognition
algorithms in a class attendance system
Jun He
College of Information
Science and Technology,
Beijing Normal University
Xinjiekouwai Street No. 19,
Beijing, China, 100875
+86 13910055846
hejun@bnu.edu.cn
Yijia Zhao
College of Information
Science and Technology,
Beijing Normal University
Xinjiekouwai Street No. 19,
Beijing, China, 100875
+86 13240290681
Yijia__zhao@163.com
Bo Sun*
College of Information
Science and Technology
Beijing Normal University
Xinjiekouwai Street No. 19,
Beijing, China, 100875
+86 13321162821
tosunbo@bnu.edu.cn
(Address all correspondence
to Bo Sun)
Lejun Yu
College of Information
Science and Technology,
Beijing Normal University
Xinjiekouwai Street No. 19,
Beijing, China, 100875
+86 13520228475
yulejun@bnu.edu.cn
ABSTRACT
Intelligent classroom is a trending topic and many
sophisticated technologies are now used in classrooms. Class
attendance is an important part of teaching management,
meanwhile video monitoring techniques and face recognition
algorithms have also made great progress. However, few researches
have focused on the applications of the latter in actual classrooms.
Aiming at an automatic class attendance system, the paper proposes
a classroom video capture scheme, whose objective is to obtain the
monitoring videos, meeting the requirements of the face
recognition algorithms. Also the paper analyzes various factors’
impacts of the face recognition, then find out the face recognition
algorithms which perform better in our experiments. Finally, the
algorithms would get the class attendance according to the video
files obtained by improved monitoring system. After the field test,
the proposed scheme could guarantee a good performance in actual
classrooms.
CCS Concepts
• CCS → Computing methodologies → Artificial
intelligence → Computer vision → Image and video
acquisition → Camera calibration
Keywords
Video monitoring system; Automatic classroom attendance; Face
recognition
1. INTRODUCTION
Class attendance is an important part of teaching management.
It is of great significance to the assessment of teaching quality. With
the rapid development of computer technology, the combination of
artificial intelligence and classrooms become more and more
popular.
The class videos obtained by the existing equipments have low
resolutions, and students’ faces in the videos have big deflection
angles. The applications of the complex monitoring networks are
only restricted to monitor students’ exams or classroom discipline.
Face recognition algorithms have made great progress, but
there are few researches focus on their applications in actual
classrooms. We think that if class monitoring system was improved,
the algorithms would get the class attendance according to the
videos. That is the inspiration of this paper.
Above all, designing a new classroom monitoring system by
using computer vision technology to get the class attendance is
necessary and innovative. To meet this requirement, we propose an
improved video capture scheme. The deflections of students’ faces
would be less than 45° and resolutions of faces would between
100*100 and 140*140 in improved videos. Then, the face
recognition algorithms could identify the students and we could get
the class attendance.
The rest of the paper is organized as follows. Section2 briefly
describes some related works. Section3 introduces some factors’
impacts on face recognition algorithms and the video capture
scheme in classrooms. Section4 first introduces the experiments
among algorithms and some factors which have huge impacts on
the performance of the face recognition, then shows the final test
result. Conclusions and the future works are included in Section 5.
2. RELATED WORKS
Nowadays, computer vision has made breakthroughs by
using deep learning technology. Many face recognition algorithms
such as Fisherface[1], SEETAFACE[2,3,4], DEEPID[5],
DEEPID2[6], DEEPID2+[7] are continually proposed. But these
algorithms are applied only under certain conditions (normal
posture, uniform light, single background), in natural condition (for
example, the light is dim, the faces in the image exist pitches or
yaws) or the object is non-cooperative, face recognition is still a
challenge. Face recognition contains three steps, face detection,
face alignment and face identification. Illumination, resolutions of
faces and deflections of faces all have impacts on face detection,
alignment and identification. However, few meticulous researches
have been done on these impacts. There are also few researches on
how to design a video capture scheme which could make all the
students’ faces in video have adequate resolutions and smaller
deflections in different-size classrooms. Some papers [8, 9, 10, 11]
propose that they could get the class attendance by installing a
camera on the front door or the front wall, the layout of the cameras
isn’t their focus.
3. VIDEO ACQUISITION SCHEME IN A
CLASSROOM
The class attendance system works as Figure. 1.
Figure. 1. The working flow of system.