SPECIAL SECTION ON GIGAPIXEL PANORAMIC VIDEO WITH VIRTUAL REALITY

Received May 18, 2020, accepted June 25, 2020, date of publication July 10, 2020, date of current version September 11, 2020.
Digital Object Identifier 10.1109/ACCESS.2020.3007205

Face Recognition Attendance System Based

on Real-Time Video Processing
HAO YANG1 AND XIAOFENG HAN 2
1 College of Media Engineering, Communication University of Zhejiang, Hangzhou 310018, China
2 College of Mathematics and Systems Science, Shandong University of Science and Technology, Qingdao 266590, China

Corresponding author: Xiaofeng Han ([email protected])

This work was supported in part by the Basic Public Welfare Research Project of Zhejiang Province under Grant LGF20H180001, and in
part by the Basic Public Welfare Research Project of Zhejiang Province under Grant LGF20F020001.

ABSTRACT With the advent of the era of big data in the world and the commercial value of face recognition
technology, the prospects for face recognition technology are very bright and have great market demand.
This article aims to design a face recognition attendance system based on real-time video processing.
This article mainly sets four directions to consider the problems: the accuracy rate of the face recognition
system in the actual check-in, the stability of the face recognition attendance system with real-time video
processing, the truancy rate of the face recognition attendance system with real-time video processing and the
interface settings of the face recognition attendance system using real-time video processing. By analyzing
the situation of these problems, the concept of attendance system based on face recognition technology
is proposed, and the research on face recognition attendance system based on real-time video processing is
carried out. Experimental data shows that the accuracy rate of the video face recognition system is up to 82%.
Compared with the traditional check-in method, the face recognition attendance system can be reduced by
about 60%. The rate of skipping classes has greatly reduced the phenomenon of students leaving early and
skipping classes. The face recognition time and attendance system with real-time video processing through
the above experimental certification can quickly complete the tasks of students in the time and attendance
check-in system, get rid of the complex naming phenomenon, greatly improve the efficiency of class, and
play an important role in guiding the development of the time and attendance system.

INDEX TERMS Video processing, face recognition technology, face recognition attendance, attendance
system, video recognition.

I. INTRODUCTION In recent years, the face recognition application system has

In this era of Internet explosion, computer technology has developed rapidly as a computer security technology in the
involved many areas of people’s lives and work. The occa- world, especially today, when terrorist activities are rampant,
sions where people come into contact with computers are this technology has received more and more attention. Face
gradually expanding. The frequency with which people use recognition technology has many typical applications in the
computing is also increasing. One of the most challenging field of public safety, civil economy, and home entertain-
projects in the field has a broad application prospect because ment [1], [2]. The pipeline of general enterprises needs to
of its huge sense of innovation. As an important identity label record the attendance of personnel, which has become a basic
for people to distinguish different individuals, face recogni- requirement of the company. However, when these attendance
tion technology has gradually entered people’s lives. Face systems are formulated, unnecessary errors often occur. Tak-
recognition is the combination of artificial intelligence and ing the current fingerprint attendance system as an example,
computer. Because of its huge challenging innovation and the study has found that the fingerprint attendance system has
broad application prospects, it has become the most challeng- an error rate of about 5%, and there will be a phenomenon
ing topic in this field. that fingerprints cannot be hit, which seriously affects the
efficiency of attendance, especially in large attendance sites,
The associate editor coordinating the review of this manuscript and which is more likely to cause congestion. However, the card
approving it for publication was Zhihan Lv . attendance system has the phenomenon of employees swiping

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
VOLUME 8, 2020 159143
 H. Yang, X. Han: Face Recognition Attendance System Based on Real-Time Video Processing

cards for someone else, and it is difficult to achieve the system time and the form of attendance system conversion
purpose of real time attendance. Compared with the two have made tremendous innovations, greatly improving the
attendance systems, the face recognition system has higher attendance rate and the reliability of face recognition technol-
accuracy and stability, because there are more points for ogy. It is worthy of further exploration and realization by our
face recognition, which is more accurate than other sys- scientists.
tems. Greatly improved, it is difficult to congestion [3], [4].
Although China’s research on face recognition technology II. PROPOSED METHOD
started late, our scientific researchers have caught up and A. REAL-TIME VIDEO FACE IMAGE RECOGNITION
some leading figures have established their own industry 1) FACE RECOGNITION
positions in the field of face recognition. With the advent of Face recognition is the core of the entire recognition process.
the era of big data in today’s world and the commercial value Face recognition is a computer vision technology that ana-
of face recognition technology, the prospect of this technol- lyzes facial feature information for identity identification. In a
ogy research is very bright and has great market demand [5]. broad sense, face recognition is divided into two parts: face
Faces in surveillance videos often suffer from serious detection and face recognition matching. Face recognition
image blur, posture changes, and occlusion. In order to over- technology is based on the facial features of the person,
come the challenges of video-based face recognition (VFR), and the input face image or video stream. First determine
Ding C has proposed a comprehensive framework based on whether there is a human face, if there is a human face,
convolutional neural network (CNN). First, in order to learn a then further give the position, size of each face and the
fuzzy and robust face representation, Ding C artificially blurs position information of each major facial organ. Based on this
the training data composed of clear still images to make up information, the identity features contained in each face are
for the lack of real video training data. Using training data further extracted and compared with known faces to identify
composed of still images and artificial fuzzy data, CNN is the identity of each face [8], [9]. Face recognition technology
encouraged to automatically learn fuzzy insensitive features. belongs to biometric recognition technology, which mainly
Second, in order to enhance the robustness of CNN features includes four parts: face image collection, face image pre-
to pose changes and occlusion, CNN has proposed a trunk processing, face image feature extraction, matching and com-
branch CNN model (TBE-CNN), which extracts complemen- bining hard recognition, combined with hardware cameras,
tarity from the overall face image and the patches around the network lines and computing device. The calculation method
face parts Information [6]. Scholars such as Nemirovskiy V is as follows:
B have studied the featureless face recognition problem. The
recognition is based on clustering the proximity between the
T = min {T1 , T2 , · · · , Tn } (1)
cardinal distributions of the luminance clusters of the divided
images. As a proximity measure, Nemirovskiy V B uses three
types of distances: Euclidean distance, cosine distance, and The principle of face recognition. Face recognition tech-
Leibler distance. Recursive neural network software model is nology is a kind of biometrics technology, through the
used for image segmentation and proximity measure cluster- acquisition of camera equipment face information and pre-
ing [7]. processing. Face detection. The main purpose of face detec-
The design of face recognition attendance system with tion is to collect information to determine whether there is
real-time video processing is conducive to the development a human face image in the image, and to determine the size
of enterprises and has a positive effect on the development and position of the image, and segment the detected human
of enterprises in the future. This article aims to design a face image into the adult face area. The last link is face
face recognition time and attendance system based on real- recognition, extracting facial feature information and image
time video processing. In this experiment, four investigation information to determine whether it is in the repository. If it
experiments were carried out: the accuracy rate of the face is, it has matching identity information, otherwise there will
recognition system in actual check-in; the stability of the be no recognition results [10]–[12].
face recognition time and attendance system with real-time
video processing; analysis of the skip rate of face recognition 2) FACE FEATURE EXTRACTION BY LDA METHOD
attendance system using real-time video processing; interface The meaning of LDA is linear discriminant analysis. This
settings of face recognition attendance system using real-time algorithm is to find a set of linear transformations that min-
video processing. The experimental results prove that the imize the intra-class dispersion between each category and
time and attendance system achieves the expected time and maximize the inter-class dispersion. Fisher is a linear judg-
attendance results through face recognition technology and ment method in LDA. The name is called because Fisher
with the help of a computer, which fully reflects the fea- function is often used for calculation. LDA can also be used
sibility design of the overall algorithm. The students who very well in face recognition, but when using this method
completed the attendance sign-in system quickly completed to extract face features, some small sample problems often
the tasks, got rid of the complicated sign of roll call, and occur. As long as a training sample of the face can be given,
soon realized the sign of operation and function. The future and the sample belongs to the other four categories, the total

159144 VOLUME 8, 2020

H. Yang, X. Han: Face Recognition Attendance System Based on Real-Time Video Processing

number of faces in the sample is:

M
X
N = M1 (2)
t=1

The average value among these categories is calculated as

follows: FIGURE 1. Main face recognition methods.

M
1 X
Z= Zt (3)
M simple unit can only solve simple calculations, but the system
t=1
composed of units in complex structures can be a complicated
The dispersion within the sample is denoted as Sw, and
question. The neural network algorithm has also achieved
the dispersion between the samples is denoted as Sb. The
good results in face recognition. For example, commonly
calculation formula is:
used BP networks, self-organizing networks, convolutional
1 t networks, etc. As long as the network is large enough, there
Sb = Mt (Zt − z) (Zt − z) = 8B 8B t (4)
N are enough training samples to theoretically recognize all
 2
8B = (Mt N ) (Zt − z) (5) faces. Although neural networks have some advantages in
8b = [8b1 , 8b2 , . . . 8bn ] (6) face recognition, they also have considerable defects. The
M M structure of neural networks is huge and complex, and their
1 XX training requires a huge sample library. The training time
Sn = (Z M − Z T ) (7)
N often takes days or even months. The speed is not fast enough.
t=1 t=1
Therefore, neural networks are not commonly used in the
3) MAIN FACE RECOGNITION METHODS actual application of face recognition.
a: GEOMETRIC FEATURE METHOD
Since the facial features such as eyes, nose, ears, mouth, etc. d: SUPPORT VECTOR MACHINE (SVM) METHOD
are different in structure, different human faces are repre- Support vector machine is a research hotspot of pattern recog-
sented according to different characteristics of the charac- nition. The basic principle of the algorithm is to use samples
teristic shapes of these organs. Geometric features were first to form a lattice in the high-level feature space, select sam-
used in the description and recognition of the side profile of ple points as the support vector near the boundary between
a human face. It determines a number of feature points based the two types of sample points, and use the support vector
on the profile line of the person’s side, and then derives a to make the decision. And finally achieve the purpose of
set of feature quantities for recognition such as angle and classification and identification. The method of projecting
distance based on these feature points. Its advantage is the into a high-dimensional space makes it possible to solve
use of simple geometric information, so the time cost of many problems that are difficult to linearly classify in the
storage space and classification is small, and it can still be status space. But support vector machines also have their
used when the image recognition rate is low; it is not sensitive shortcomings. First, support vector machines are a two-class
to changes in lighting. Its disadvantage is that it is difficult to classification algorithm. Although some methods can be
extract stable features from the image, it is greatly affected used to solve multi-classification problems, their efficiency
by changes in posture and expression, and the stability is not is often greatly reduced. Secondly, support vector machines
high [13]. want high-dimensional space projection, which requires the
support of kernel functions, but choosing kernel functions
b: SUBSPACE ANALYSIS METHOD is indeed a lot of trouble. Finally, although support vector
The analysis method of subspace is to use spatial transforma- machines can classify independently, the effect of directly
tion to map face image data into a certain subspace to achieve classifying faces is not good, and often feature extraction is
dimensionality reduction of face data. It is a huge face data required for faces.
that becomes simple to calculate, and then to the dimensional- The dimensionality reduction process is performed before
ity reduction data classification. Different subspace analysis recognition, and the basic flow is shown in Figure 1 [7].
methods use different criteria, and different subspaces can
be obtained. Common subspace analysis methods for face 4) VIDEO IMAGE RECOGNITION SYSTEM
recognition include: principal component analysis, linear dis- The face detection technology is used to locate and segment
criminant analysis, independent element analysis, etc. a partial face image from the image; the feature extraction
technology extracts the amount of data that can characterize
c: NEURAL NETWORK METHOD the face image, and forms the features to be stored in the
Neural network is a commonly used method in membrane feature database. Face recognition process: face positioning
recognition. Its principle is to use a large number of simple and image processing in the image; feature extraction and
calculation units to form a certain hierarchical structure. Each selection; detecting and recognizing the image, and returning

VOLUME 8, 2020 159145

 H. Yang, X. Han: Face Recognition Attendance System Based on Real-Time Video Processing

the recognition result. First create a facial image file of the To sum up, the system selects Gabor features plus Fisher
face. That is, use the camera to collect facial image files of the based discriminant analysis method based on orthogonal
person’s face or take their photos to form a facial image file, basis to become a linear discrimination method [17], [18].
and store these facial image files to generate faceprint codes. Face image feature representation is a key part in face
Get the current human face. That is, use the current facial recognition, and good feature representation can improve the
image captured by the camera, or take a photo input, and robustness of image matching. Gabor wavelet feature descrip-
generate a facial texture code from the current facial image tion method is a comprehensive method that combines gray
file. Compare with the current facial texture encoding and file and local descriptions. It has the advantages of gray-based
inventory. That is, to retrieve and compare the current facial and feature-based methods [19].
texture code with the facial texture code in the file inventory. Gabor kernel definition of wavelet transform:
The above-mentioned ‘‘face coding’’ method works accord- kEj2 kEj2 xE2 σ2
ing to the essential characteristics of the human face and the 9j (Ex ) = 2 exp(− 2 )[exp(ikEj xE) − exp(− )] (8)
beginning. This facial coding can resist changes in light, skin σ 2σ 2
tone, facial hair, hairstyle, glasses, expression and posture, E
xE = (x, y) is a spatial domain variable, and kj is a frequency
and has strong reliability, so that it can accurately identify a vector that determines the scale and direction of the Gabor
person from millions of people. The face recognition process kernel. We selected an image I (Ex ), and the Gabor transform
can be completed automatically, continuously, and in real at the xE0 point at a specific position in the picture can be
time using ordinary image processing equipment [14]. The realized by convolution with Gabor kernel at this point. The
video image recognition system is mainly composed of four face image is I (z), and the z = (x, y) picture scale direction is
parts: login module, recognition module, check-in module µ, ν, there is a formula:
and background management module [11]. Taking the school
system as an example, the functions of each module are as Gµ,ν (z) = I (z) ∗ ψµ,ν (z) (9)
follows: With z = (x, y), the calculation formula can be obtained:
1) The login module is where the lecturer or background
administrator logs in with an account and password to µ2 + ν 2 (µ2 + ν 2 )(x 2 + y2 )
G(x, y) = exp(− )
view attendance information. σ2 2σ 2
2) The main function of the recognition module is to σ2
receive a face picture, call the system application pro- × [exp(i(µx + νy)) − exp(− )] (10)
2
gramming interface (Application Programming Inter-
face, API) to perform face recognition on the picture, Among them, k = ( µν ), µ = kfmax πM
N cos( 8 ), υ =
πM π
√
and obtain an identification code that uniquely deter-
kmax
fN
sin( 8 ), kmax = 2 , f = 2, σ = 2π
mines the picture. Fisher discriminant analysis is an improved algorithm
3) The check-in module receives the identification code based on PCA. It adopts the method of dimensionality reduc-
obtained in the identification module and compares tion to effectively reduce the amount of calculation, thereby
it with the student information in the database to improving the calculation speed [20]. This method can ensure
find and confirm the student information; by querying that the projected samples have the smallest intra-class dis-
the current time and the schedule information in the tance and the largest inter-class distance in the new space,
database, the current course information is obtained. that is, the best separability in the space. Introduce Fisher
After obtaining the student and course information, it is discriminant criteria table:
logical to determine whether the student has checked in ϕ T Sb ϕ
for this class, if it has been checked in, ignore it; if it Ifisher (ϕ) = (11)
ϕ T Sw ϕ
has not been checked in, add the check in information
to the database check in table. The module will return where ϕ is any n-dimensional column vector. Fisher’s linear
check-in information, including: whether the check-in discrimination method selects the vector ϕ that maximizes
is successful; the student’s name; the student’s student Ifisher (ϕ) as the projection direction, so that the projected
ID; the course that is checked in; if the check-in is samples have the largest inter-class dispersion and the small-
unsuccessful, what is the reason. est intra-class dispersion. Among them, Sb is the inter-class
4) The background management module is responsible for separation matrix, and Sw is the intra-class separation matrix.
the background administrator, which mainly includes Let Fk0 , k = 1, 2, . . . , L be the average of the training samples
the functions of counting sign-in information, entering of class ϕk after PCA and LDA transformation. Use the
or deleting student information, etc. nearest domain classifier for face recognition classification:
δ(Y , Fk0 ) = min δ(Y , Fk0 ), Y ∈ ϕk (12)
B. BASIC FACE RECOGNITION ALGORITHM
When designing a system to select a face recognition algo- Among them, δ is judged as the distance. If the distance
rithm, we have to consider the following factors: recognition between Y and Fk0 is smaller than the distance between the
rate, algorithm robustness, and matching time [12], [15], [16]. two face images in class Fk0 , then Y and Fk0 can be regarded

159146 VOLUME 8, 2020

H. Yang, X. Han: Face Recognition Attendance System Based on Real-Time Video Processing

the face recognition attendance system using real-time

video processing.
(2) The stability of face recognition attendance system
with real-time video processing
Taking the manual punch card as the control group,
the face recognition attendance system using real-time
FIGURE 2. Complete attendance system module.
video processing is used for the check-in statistics, and
the number of check-in punch cards is used as the main
as a class of pictures, and the recognition is successful; oth-
observation parameter to detect the stability of the face
erwise, they are not classified as similar pictures and Y is a
recognition attendance system.
non-training sample image.
(3) Analysis of the truancy rate of face recognition atten-
dance system using real-time video processing
C. BASIC MODULE OF FACE RECOGNITION ATTENDANCE
Taking the manual punch card as the control group,
SYSTEM BASED ON REAL-TIME VIDEO PROCESSING
and using the real-time video processing face recog-
The complete attendance system consists of a combination nition attendance system to perform real-time check-in
of multiple modules, each of which performs different func- statistics, observe the students’ skip-rate rate of the two
tions. In order to reduce the complexity of the program and methods.
facilitate the reuse of codes, it is conducive to maintain and (4) Interface settings of face recognition attendance system
manage the entire system. The design of the face recognition using real-time video processing
time and attendance system in this system mainly includes By identifying the student’s face as the attendance
several modules of video acquisition terminal, cable trans- interface, the successful student attendance and time
mission module, data storage, face recognition module and information will be displayed on the screen. By select-
computer terminal module [21], [22], as shown in flow chart2. ing face detection and recognition, the video images of
the attendance students can be described by the face
III. EXPERIMENTS recognition attendance system multi-person video.
A. EXPERIMENTAL SETUP
1) EXPERIMENTAL BACKGROUND C. DATABASE DESIGN
In order to deeply study the application of face recognition This system uses mysql database for data storage. MySQL
attendance system for real-time video processing, from the has the advantages of fast speed, low cost, small size, open
accuracy of actual recognition of face recognition system, source, support for multi-threading and multi-processor, good
the stability of background application system of face recog- portability, and at the same time can use ODBC (open
nition attendance system, the face recognition system faces database connection) and SQL (structured query language)
some problems; the development status of the role played by to increase the data (create), delete (delete), query (select),
the face recognition system in the actual check-in, the factors modify (update) operations.
restricting the development, and the difficulties faced by the D. FACE RECOGNITION MODULE DESIGN
development are considered, and many problems are found. This system uses a variety of popular programming
languages-Python (Interpretive Language), Java, C ++ and
2) EXPERIMENT SETUP PROCESS other languages to develop. Python is used for real-time video
The experiment sets up a control group and an experimental data collection and face recognition processing. C ++ is
group, the control group uses traditional fingerprint check-in, used to complete the related system running components such
and the experimental group uses a real-time video processing as file operations and client running interface design. Java
face recognition attendance system. Choose two universi- is used to build a face recognition WEB platform service.
ties from a province, choose the same number of students Combined with the third-party library Open CV to realize the
for the experiment, collect, count and analyze the experi- writing of face recognition module, OpenCV provides rich
ment, analyze the application space, development prospects visual processing and image processing algorithms to analyze
of the face recognition system in the actual check-in, and the and process the facial features captured by the classroom
face recognition system in some problems with the actual camera. The system is divided into three modules: front-end
check-in. operation, real-time video face recognition module, and back-
ground data management.
B. EXPERIMENTAL PROCEDURE
(1) Accuracy rate of face recognition system in actual IV. DISCUSSION
check-in A. REAL-TIME VIDEO PROCESSING FACE RECOGNITION
The face recognition attendance system using real-time ATTENDANCE SYSTEM IN THE ACTUAL SIGN-IN
video processing is used to count the sign-in rates ACCURACY RATE
of students from two colleges and universities, and By investigating two colleges A and B in a province, we con-
to compare and analyze the sign-on accuracy rate of ducted experiments in colleges based on the application of

VOLUME 8, 2020 159147

 H. Yang, X. Han: Face Recognition Attendance System Based on Real-Time Video Processing

TABLE 1. Video face recognition accuracy. TABLE 2. Stability analysis of face recognition attendance system based
on real-time video processing.

FIGURE 3. Video face recognition accuracy.

face recognition attendance system accuracy, and selected

200 college students who need to punch cards. The statistical
results of the survey are shown in Table 1 and Figure 3.
According to the experimental results, the accuracy rate of
face recognition in the classroom video of the two colleges
is currently high, and the accuracy rate of face recognition FIGURE 4. Stability analysis of face recognition attendance system based
on real-time video processing.
in the college video is about 82%. About 15% of the failed
card punches due to video blur and other reasons, About 3%
of the staff failed to punch in, it can be seen from these data TABLE 3. Statistical table for checking student attendance.
that the accuracy rate of the video face recognition system is
relatively high.

B. STABILITY ANALYSIS OF FACE RECOGNITION

ATTENDANCE SYSTEM BASED ON REAL-TIME VIDEO
PROCESSING online in real time. Compared with the traditional finger-
The face recognition attendance system for real-time video print check-in, students’ rate of skipping classes will change.
processing performs video punching every two hours. Com- The survey data is shown in Table 3 and Figure 5. The
paring the data of the face recognition attendance system results showed that after using the face recognition system,
and manual fingerprint punching, the sign-in success is the skipping rate of the two universities was significantly
1 and the failure is 0. The data collection results are shown lower than that of the control group, only about 13%. There-
in Table 2 and Figure 4. An error occurs in the face recogni- fore, the face recognition attendance system can effectively
tion attendance system from 7 am to 9 am; from 13 noon to increase the attendance rate of university classrooms, and
21 pm, the face recognition attendance system and manual is an effective method to restrain students from skipping
fingerprint punch card can correctly identify the check-in. classes.
The experimental results show that the length of time when
the smart machine is turned on may bring some experimental D. FACE RECOGNITION ATTENDANCE SYSTEM INTERFACE
errors, indicating that the human-machine interactive testing SETTINGS
instrument needs to be pre-powered on for two to four hours From the amount of change in Table 4 and Figure 6, it can be
before the accuracy of the testing data can be guaranteed. seen that by recognizing the student’s face as the attendance
interface, the successful attendance and time information of
C. ANALYSIS OF THE SKIPPING RATE OF FACE the confirmed student is displayed on the screen. By select-
RECOGNITION ATTENDANCE SYSTEM BASED ON ing face detection and recognition, the video images of the
REAL-TIME VIDEO PROCESSING attendance students can be described by the face recognition
Because the face recognition attendance system has a attendance system multi-person video. Compared with the
real-time video function, it can count the number of people traditional punch card signing, this method greatly improves

159148 VOLUME 8, 2020

H. Yang, X. Han: Face Recognition Attendance System Based on Real-Time Video Processing

V. CONCLUSION
(1) College attendance management for students has
become one of the hot issues in the society, so the
management of college students should be strength-
ened. However, most college students still use tradi-
tional manual attendance for daily attendance, using
paper signatures or teacher orders, but now with the
gradual rise of technology, some new methods point
out that gradually, a few colleges and universities
FIGURE 5. Statistics chart for checking student attendance. will use punch card fingerprints and smart atten-
dance methods. Although there are some ways to
TABLE 4. Test number and accuracy rate table.
stimulate attendance, the effect is not so effective.
Attendance and these methods have a common short-
coming, fraud will occur, thereby increasing the rate
of absenteeism. This repeated phenomenon not only
has a negative impact on students’ psychology and
physiology, but also maintains the normal order of
university teaching and hinders the quality of teaching.
At the same time it will have a very unfavorable situa-
tion training and the formation of university spirit and
discipline.
(2) In this article, a face recognition attendance system
based on real-time video processing is designed, and
two colleges in a province are selected for real-time
check-in and inspection of student attendance. This
article mainly sets four directions to consider the prob-
lems: the accuracy rate of the face recognition sys-
tem in the actual check-in, the stability of the face
recognition attendance system with real-time video
processing, and the truancy rate of the face recognition
attendance system with real-time video processing It
is difficult to analyze the interface settings of the face
recognition attendance system using real-time video
processing. By analyzing the situation of these prob-
lems, the concept of attendance system based on face
FIGURE 6. Test number and accuracy rate chart. recognition technology is proposed, and the research on
face recognition attendance system based on real-time
video processing is carried out. Research data shows
the efficiency, and can prevent early leave and skip class. that the accuracy of the video face recognition system
By regularly extracting images, seeing that the students ’han- is about 82%. The face recognition time attendance
dling of the case is transparent, the inspection method is an system and manual fingerprint punching are more sta-
imperceptible process to a process where the user perceives ble and correctly identify check-ins, and the rate of
no signs, it is a sign of an automated way. After repeated test- skipping classes is significantly reduced compared with
ing, the test results of the system under different test sample the control group, only about 13%. Compared with the
numbers are as follows: when the input student is 10 ∼ 20, control group, the efficiency is greatly improved, which
the recognition and sign-in correct rate is 85%; when the can prevent students from leaving early and skipping
input student is 20 ∼ 50, the correct rate is 80%. There is classes.
an error rate of about 20% in the test results. Most students (3) The attendance system realizes the expected attendance
face information errors that will change, such as changes in results through face recognition technology with the
facial features, accessories, cosmetics and lighting caused by help of a computer, which fully reflects the feasibil-
medical plastics, making it impossible to extract the correct ity design of the overall algorithm. The students who
logo from the picture; Another main reason is the accuracy have completed the attendance sign-in system quickly
of identification code comparison. The system fails to cor- completed the tasks, got rid of the complicated sign of
rectly judge that the two identifiers are very similar. Under roll call, and soon realized the sign of operation and
various perfect conditions, make sure to adapt to various function. The system has made tremendous innova-
situations. tions, greatly improving the attendance rate and the

VOLUME 8, 2020 159149

 H. Yang, X. Han: Face Recognition Attendance System Based on Real-Time Video Processing

reliability of face recognition technology. It is worthy [17] Y. Li, W. Song, and C. Cheng, ‘‘Attendance system of face recognition
of further exploration and realization by our scientists. based on raspberry pi%’’ Microcontrollers Embedded Syst. Appl., vol. 16,
no. 11, pp. 28–30, 34, 2016.
[18] S. Bharadwaj, H. S. Bhatt, M. Vatsa, and R. Singh, ‘‘Domain specific learn-
REFERENCES ing for newborn face recognition,’’ IEEE Trans. Inf. Forensics Security,
[1] K. Solanki and P. Pittalia, ‘‘Review of face recognition techniques,’’ Int. vol. 11, no. 7, pp. 1630–1641, Jul. 2016.
J. Comput. Appl., vol. 133, no. 12, pp. 20–24, Jan. 2016. [19] X. Zhu, H. Liu, Z. Lei, H. Shi, F. Yang, D. Yi, G. Qi, and S. Z. Li, ‘‘Large-
[2] C. Stoll, R. Palluel-Germain, R. Caldara, J. Lao, M. W. G. Dye, F. Aptel, scale bisample learning on ID versus spot face recognition,’’ Int. J. Comput.
and O. Pascalis, ‘‘Face recognition is shaped by the use of sign language,’’ Vis., vol. 127, nos. 6–7, pp. 684–700, Jun. 2019.
J. Deaf Stud. Deaf Educ., vol. 23, no. 1, pp. 1–9, 2018. [20] H. Nguyen, W. Yang, B. Sheng, and C. Sun, ‘‘Discriminative low-rank
[3] W. Deng, J. Hu, and J. Guo, ‘‘Face recognition via collaborative representa- dictionary learning for face recognition,’’ Neurocomputing, vol. 173,
tion: Its discriminant nature and superposed representation,’’ IEEE Trans. pp. 541–551, Jan. 2016.
Pattern Anal. Mach. Intell., vol. 40, no. 10, pp. 2513–2521, Oct. 2018. [21] H.-K. Ji, Q.-S. Sun, Z.-X. Ji, Y.-H. Yuan, and G.-Q. Zhang, ‘‘Collaborative
[4] T. Pei, L. Zhang, B. Wang, F. Li, and Z. Zhang, ‘‘Decision pyramid classi- probabilistic labels for face recognition from single sample per person,’’
fier for face recognition under complex variations using single sample per Pattern Recognit., vol. 62, pp. 125–134, Feb. 2017.
person,’’ Pattern Recognit., vol. 64, pp. 305–313, Apr. 2017. [22] L. Wang and S. Chen, ‘‘Joint representation classification for collective
[5] H. Shi, X. Wang, D. Yi, Z. Lei, X. Zhu, and S. Z. Li, ‘‘Cross-modality face face recognition,’’ Pattern Recognit., vol. 63, pp. 182–192, Mar. 2017.
recognition via heterogeneous joint Bayesian,’’ IEEE Signal Process. Lett.,
vol. 24, no. 1, pp. 81–85, Jan. 2017.
[6] C. Ding and D. Tao, ‘‘Trunk-branch ensemble convolutional neural net-
works for video-based face recognition,’’ IEEE Trans. Pattern Anal. Mach.
Intell., vol. 40, no. 4, pp. 1002–1014, Apr. 2018.
[7] V. B. Nemirovskiy, A. K. Stoyanov, and D. S. Goremykina, ‘‘Face recog-
nition based on the proximity measure clustering,’’ Inst. Cybern. Tomsk
Polytech. Univ., vol. 40, no. 5, pp. 740–745, 2016. HAO YANG was born in Yongkang, Zhejiang,
[8] K. Taniya, M. Nidhi, and T. Nandini, ‘‘Automated human resource and China, in 1986. He received the Ph.D. degree from
attendance management system based on real time face recognition,’’ the University of Chinese Academy of Sciences,
IJSRSET, vol. 16, no. 4, pp. 847–853, 2016. China. He currently works with the College of
[9] D. Wu, Y. Tang, G. Lin, and H. Hu, ‘‘Roboust face recognition based on Media Engineering, Communication University of
significance local directional pattern and deep learning,’’ J. Optoelectron. Zhejiang. His research interests include formal
Laser, vol. 27, no. 6, pp. 655–661, 2016. method and face recognition.
[10] Y. Sun, J. Zhao, and Y. Hu, ‘‘Supervised sparsity preserving projections for
face recognition,’’ Proc. SPIE, vol. 8009, no. 4, pp. 357–366, 2017.
[11] L. Best-Rowden and A. K. Jain, ‘‘Longitudinal study of automatic face
recognition,’’ IEEE Trans. Pattern Anal. Mach. Intell., vol. 40, no. 1,
pp. 148–162, Jan. 2018.
[12] T. Valentine, M. B. Lewis, and P. J. Hills, ‘‘Face-space: A unifying concept
in face recognition research,’’ Quart. J. Experim. Psychol., vol. 69, no. 10,
pp. 1996–2019, Oct. 2016.
[13] Y. Duan, J. Lu, J. Feng, and J. Zhou, ‘‘Context-aware local binary feature
learning for face recognition,’’ IEEE Trans. Pattern Anal. Mach. Intell.,
XIAOFENG HAN was born in Taian, Shandong,
vol. 40, no. 5, pp. 1139–1153, May 2018.
[14] S. Chakraborty, S. K. Singh, and P. Chakraborty, ‘‘Local gradient hexa China, in 1982. He received the master‘s degree
pattern: A descriptor for face recognition and retrieval,’’ IEEE Trans. from the Shandong University of Science and
Circuits Syst. Video Technol., vol. 28, no. 1, pp. 171–180, Jan. 2018. Technology, China. He currently works with the
[15] Q. Feng, C. Yuan, J.-S. Pan, J.-F. Yang, Y.-T. Chou, Y. Zhou, and W. Li, College of Mathematics and Systems Science,
‘‘Superimposed sparse parameter classifiers for face recognition,’’ IEEE Shandong University of Science and Technology.
Trans. Cybern., vol. 47, no. 2, pp. 378–390, Feb. 2017. His research interests include image processing,
[16] S. P. Mudunuri and S. Biswas, ‘‘Low resolution face recognition across information, and computing science.
variations in pose and illumination,’’ IEEE Trans. Pattern Anal. Mach.
Intell., vol. 38, no. 5, pp. 1034–1040, May 2016.

159150 VOLUME 8, 2020