International Journal of Interactive Mobile Technologies(iJIM) – eISSN: 1865-7923 – Vol. 16 No. 08 (2022)


Paper—Automated Cheating Detection based on Video Surveillance in the Examination Classes

Automated Cheating Detection based on Video 
Surveillance in the Examination Classes

https://doi.org/10.3991/ijim.v16i08.30157

Roa’a M. Al_airaji1(*), Ibtisam A. Aljazaery2, Haider Th. Salim ALRikabi3, 
Abdul Hadi M. Alaidi3

1Information Technology College, Babylon University, Babylon, Iraq 
2Engineering College, Babylon University, Babylon, Iraq

3Electrical Engineering Department, College of Engineering, Wasit University, Wasit, Iraq
roaaalairaji90@gmail.com

Abstract—A major issue in the field of education is exam malpractice and 
other forms of academic dishonesty, where efforts made towards the assessment 
of students are undercut. In order to address this menace, a cheating detection 
system has been proposed in this study. The system can detect cheating in real-
time. The proposed system uses video surveillance to monitor the activities of 
the student during exams, especially abnormal behavior. The development of the 
system involved the use of three different techniques, with different functions. 
The first technique helps in determining the direction of the students’ heads when 
they move from their initial direction, which is the exam script. Some form of 
cheating involves students peeping at the exam scripts of other people writing 
the same exam. Whenever the system observes a deviation that exceeds the set 
threshold, it classifies the behavior as abnormal. The second technique works 
with the movement of the student’s iris. It detects when a student’s iris moves in 
a different direction to copy answers from written documents like mobile phones, 
books, hands, summary papers. The third technique is used in identifying the 
contact between a student’s hands and face, as well as that between different stu-
dents for shared abnormal behavior detection between students, such as sharing 
of incriminating materials. When any of these is detected, an automatic alarm 
alerts the authority about the abnormal behavior that has been detected, thereby 
minimizing the error rate that can occur as a result of manual monitoring.

Keywords—surveillance videos, videos analysis, behaviors recognition, 
anomaly detection, neural networks

1 Introduction

Surveillance is the process through which a behavior, activity, or unusual behav-
ior is monitored with the aim of directing, protecting, managing, or influencing them 
[1, 2]. In government agencies, the use of surveillance is employed in crime investiga-
tion, gathering of intelligence, or safeguarding a procedure, person, or object. Effective 

124 http://www.i-jim.org

https://doi.org/10.3991/ijim.v16i08.30157
mailto:roaaalairaji90@gmail.com


Paper—Automated Cheating Detection based on Video Surveillance in the Examination Classes

video systems designed for surveillance work by detecting suspicious or unusual 
behavior. Recently, the use of these systems has proven to be efficient in the detection 
of suspicious or unusual behavior, which in turn enables the prevention of security 
breaches [3, 4].

The faces of humans as well as their patterns of behavior are crucial to the process 
of identifying an individual, which mainly relies on visual details. The details of an 
individual’s physical appearance are obtained using the surveillance video which can 
be played back after it has been recorded or can even be recorded for live viewing as 
the activity is being recorded. The influence of the modern tendency of ‘automation’ is 
felt in every sphere of life, including the area of video analytics [5, 6]. There is a wide 
variety of domains in which the use of video analytics can be employed, including iden-
tification of humans, prediction of human behavior, recognition of abnormal activity, 
detection of motion, vehicle counting, people counting at crowded places, etc. [7]. Two 
ways that a person can be identified are through recognition of gait and faces. The rec-
ognition of faces varies in terms of automatic recognition through video surveillance. 
The use of face recognition can be employed in predicting the head orientation of a per-
son, which can, in turn, enable the prediction of a person’s behavior. A combination of 
movement detection and face detection can be effective in producing desired results in 
several applications like person verification, identification of individuals, and detection 
of a person’s presence or absence at a given location and time [8]. More so, an iden-
tification and recognition system for the detection of unusual activity among students 
writing exams can be developed based on estimation, subtle contact, detection of head 
motion, recognition of hand gestures.

The field of education has experienced academic dishonesty, especially cheating 
during exams as a big challenge [9, 10]. Normally, in an offline exam setting, stu-
dents are monitored manually by human invigilators and surveillance videos globally. 
However, monitoring an exam hall in this manner can be prone to errors and may not 
produce the best results. The attention of a human invigilator can be diverted, thereby 
reducing the focus of the invigilator on all students; this can give room for cheating 
during an examination [11, 12]. Manual supervision is a very challenging task as it 
requires sufficient and efficient manpower to carry out the work effectively. However, 
this is usually not possible to achieve, and this makes it important to have a video sur-
veillance system through which suspicious and abnormal behaviors can be automati-
cally identified and minimized suspicious and abnormal behaviors during exams. More 
so, with an automated video surveillance system, the possibility of inaccuracy can be 
minimized to a large extent, while the efficiency of an academic examination system 
can be increased [13]. In this paper, a real video of students captured while writing an 
exam in an exam hall is used in analyzing the activities and behaviors of the students. 
With this, the captured behavior of students can be categorized as abnormal or nor-
mal. The approach which is proposed in this paper is capable of recognizing abnormal 
actions and motions of the iris and head, which in turn prevents students from involving 
in exam malpractice. Also, the proposed system is capable of detecting position swap-
ping by students or leaving a position. Lastly, with the system, contacts made among 
students can be detected, thus, preventing the exchange of incriminating materials.

iJIM ‒ Vol. 16, No. 08, 2022 125



Paper—Automated Cheating Detection based on Video Surveillance in the Examination Classes

2 Research method 

Figure 1 below shows the proposed system, which is made up of different stages, 
starting with the preprocessing step that is made up of three sub-stages (splitting of 
input video into frames, elimination of noise and detection of an object in motion, 
then the introduction of the pre-processed frames into the system. Also, this system 
makes use of three key methods for the detection of movement, which are, detection of 
iris movement, detection of head movement, detection of hand movement. Lastly, the 
behavior detected by the system is classified as either abnormal or normal. As output, 
the system raises an alarm to notify the authority of the abnormal behavior that has 
been detected. 

Fig. 1. Student behavior detection system

2.1 Pre-processing

At the beginning of the pre-processing, the input video is split into frames and then 
the noise is removed through the use of Gaussian filters. After these processes, the next 
is the detection of moving objects, and this stage is made up of two sub-stages (mod-
eling and background subtraction. The outcome of the detection of objects in motion 
is foreground frames that are made up of moving objects. There are two Backgrounds. 
Firstly, the Background model refers to the background image that contains just static 
objects and the environment in the video. A background model can be derived through 
two ways, which are Initialization of background model and maintenance of back-
ground model. In the first step, the median algorithm is used; the efficiency of this algo-
rithm is hindered when the moving objects remain static for 60% of the period required 

126 http://www.i-jim.org



Paper—Automated Cheating Detection based on Video Surveillance in the Examination Classes

for training, during which the addition of the moving objects is made to the background 
image, thereby yielding incorrect intensity values. A more efficient algorithm produces 
the Highest Redundancy Ratio (HRR) as it can create background images by consid-
ering redundant pixels during the training stage, therefore overcoming the aforemen-
tioned problem of incorrect intensity values [14]. In the second step, the background 
is updated constantly due to the addition of certain external factors (such as minor 
displacement of the camera, little modifications made to the background, changes in 
lighting) to the background, especially when the motion objects are unfailingly in a 
static state. These are an integration of the modifications with the present background 
image. In this regard, there are two adaptations that the system follows; sequential 
and periodical adaptations. Sequential adaptation involves the use of a mathematical 
background model, where a low-pass filter is applied to each pixel. The statistical back-
ground model also provides the mechanism which allows the system to adapt to minor 
changes made in the background. On the other hand, the use of periodical adaptation 
is employed when there is a need to synchronize lighting with other physical changes 
that may be made to the background such as the removal or addition of objects in the 
system [15, 16]. Secondly, Background Subtraction: method of adaptive background 
subtraction refers to the approach that is employed to detect the foreground objects 
from the n image. To get the foreground image, Equation (1) below can be used. Where 
In(x) denotes the current value of intensity at pixel x and Bd(x) indicates the current 
background intensity value at pixel x, then x represents the foreground pixel if the 
equation (1) below is derived [14, 17].

 In x Bd x Tn x( ) ( ) ( )� �  (1)[8]

While the initial set to an experimental value of non-zero value which is denoted as 
sT0(x) and the first frame’s initial is represented by Bd0(x), together T0s(x) and Bd0(x) 
are constantly updated.

2.2 Detection head movement 

Subsequent to the pre-processing, the next step involves the detection of head move-
ment. The head detection technique helps in determining the direction of the student’s 
head, especially when it deviates from the answer script to a different direction. This 
technique is particularly efficient in a situation whereby, students are peeping at other 
students’ answer scripts. If the behavior goes beyond the set threshold, then the sys-
tem classifies the behavior as abnormal. There are two sub-phases involved in head 
detection. 

Face detection. In this work, the use of Viola-Jones Algorithm is employed in the 
detection of the face so that high accuracy can be achieved. Also, the features of the 
face were extracted using Gabor filters, while PCA was used for the reduction of data 
dimension. Lastly, the classification was done using a backpropagation neural network. 

iJIM ‒ Vol. 16, No. 08, 2022 127



Paper—Automated Cheating Detection based on Video Surveillance in the Examination Classes

Face recognition with Viola and Jones algorithm. In this study, the detection of 
the face has been achieved using the Viola-Jones algorithm. The preference for this 
algorithm is based on the fact that it can run in real-time with a high rate of detection 
accuracy. As output, the algorithm separates faces from non-faces from the original 
image. In other words, the algorithm extracts the image of faces from the authentic 
image. There are three major stages in Viola and Jones’s algorithm, which involves the 
use of three algorithms including, integral image extraction and its features, cascade 
classifier, and Ada Boost. The Integral Image refers to an algorithm that is cost-effective 
in generating the total intensity values of a pixel within a given rectangular area of 
an image. Using this algorithm, the Haar structures that are used in the Viola-Jones 
detector can be calculated speedily. Equation (1) below is used in calculating integral 
images. In an integral image, the position of ( x, y) is calculated by summing the pixels 
to the above and left of that location [18].

 G x y I x y
x x y y

( , ) ( , )
,

� � �
� �� �
�  (2)[16]

Where I denote the original image, while the integral image is represented by G. One 
of the most efficient ways of calculating the total of any rectangle region is the use of 
the integral image as presented in Figure 2. In computing all the pixels in rectangle 
ABCD, only four values are wanted from the essential image.

Fig. 2. Fundamental image and Haar features (sixe kinds)

In this paper, two kinds of Haar-like features are extracted for every image sample: 
the perpendicular feature and the flat feature (kind (a, b) in Figure 2). Here the total of 
the white pixels is subtracted from the total of black pixels to enable the computation 
of a Haar-like feature value. In the case whereby the value of the Haar-like feature is 
beyond the given threshold, then it is pointing to a part of the face like nose, eyes, and 
cheeks. AdaBoost is an algorithm of learning classification that is deployed in both the 
selection of a few features and training of the classifier [19, 20]. It is used when there 
is a need to compute a large number of Haar features for each rectangular area. To 
reduce the redundancy of the features, the use of Ada boost is employed. The algorithm 
achieves this by determining and eliminating the insignificant features, followed by the 
conversion of the huge amount of significant features into consolidated one. Ada boost 
is a multifaceted algorithm that is created using the weighted combination of weak 

128 http://www.i-jim.org



Paper—Automated Cheating Detection based on Video Surveillance in the Examination Classes

classifiers. Its robustness lies in its ability to choose the best features from the extant 
features, and as such, it can be regarded as a feature selection algorithm. Ada boost 
can choose features that are important enough for face characterization [17]. Cascade 
classifier is made up of different steps, containing a strong classifier from AdaBoost. At 
every phase, the status of an image is determined (whether or not it is face). If during 
a given phase, the algorithm classifies a sub-window as a face, it is then moved to the 
subsequent phase in the cascade. If classified as otherwise, then the image is discarded 
immediately [21, 22].

Features extraction using Gabor filters. This sub-phase involves the extraction of 
features from the images, which may be images of faces or non-faces. Subsequently, 
the features vector that has been selected is then moved to the next phase. In this 
work, the use of Gabor wavelets filters has been employed in the extraction of features. 
The use of Gabor filters transform has been employed in this study for two purposes, 
which are to enhance the orientation and resolution properties of the image. It is also 
an efficient method of measuring local spatial frequency. Gabor filters are one of the 
most effective ways to separate and represent texture because the manner in which it 
performs these tasks is similar to that of human visual perception. A 2D Gabor filter 
represents Gaussian functions that are created through sinusoidal waves. The reason 
for the use of the 2D Gabor wavelet filters is to acquire features vector which is meant 
to be used as input in the next stage of the system; this will help in producing the most 
significant features of the face image [23, 24]. The coefficients of Gabor consist of 
both real and imaginary parts that are created through the use of a complex function 
that combines both, or the two could be used independently as presented in the three 
equations below [25].

The complex part: 

 g x y exp
x y

exp i
x

( , , , , , , )� � � �
�
�

�
�

� �� �
� � ��

�
�

�
�
�

�
�

�
�
�

�
�
�

�

�

2 2 2

22
2��

�

�
�  (3)[12]

The imaginary part: 

 g x y exp
x y

sin
x

I ( ), , , , , ,� � � �
�
�

�
�

� �� �
� � ��

�
�

�
�
�

�
�

�
�
�

�
�
�

2 2 2

22
2  (4)[12]

The real part: 

 g x y exp
x y

cos
x

R ( ), , , , , ,� � � �
�
�

�
�

� �� �
� � ��

�
�

�
�
�

�
�

�
�
�

�
�
�

2 2 2

22
2  (5)[12]

Where: xʹ = x cosθ + y sinθ, yʹ = x sinθ + y cosθ, γ: Spatial aspect ratio, σ: Sigma, ψ: 
Phase offset, θ: Orientation angel, λ: Wave length. 

3 Reduction of features using PCA

Subsequent to the extraction of features from the image using the Gabor filter, the 
next step is to reduce the features so that the processes of converging and training 
will be completed within a reasonable time frame. To achieve timely processing of the 

iJIM ‒ Vol. 16, No. 08, 2022 129



Paper—Automated Cheating Detection based on Video Surveillance in the Examination Classes

features, the features must be reduced as the NN classifier cannot be fed with the large 
number of features that need to be classified as either face or non-face, because this will 
consume a lot of time. Thus, the use of PCA is employed in reducing the dimensions of 
the features which is the input dataset. This involves the removal of redundant or irrel-
evant details from the data and retaining the relevant ones. This way, the complexity of 
the process and time required for the detection of the face is reduced, thereby enhancing 
the optimal performance of the classifier [26].

The PCA works by reducing the high dimensionality of the input data to low dimen-
sions. Thus, it is possible to reduce the dimensions of the input data to N where (N less 
than M) when the PCA algorithm is used for a particular vector x of M dimension and 
a correlation matrix R, as shown down [27].

1. Calculate the eigenvalues λi and their corresponding eigenvectors Qi of the correla-
tion matrix R:

 R QiQi i� ��  (6)

2. The eigenvalues are reordered in descendant order: 

	 λ1 > λ2 > … > λN > … > λM

3. The eigenvector of the first N highest eigenvalues is selected. 
4. The compacted vector C is calculated by: 

 C x Qi
T

i= � � � � �for  i = 1, 2, ..., N.  (7)

4 Arrangement via neural network (NN)

The database which was used in this paper is made up of several faces and non-face 
images. The Neural Network is trained to classify face/non-face images by using the 
extracted features of the images, whose dimensions have been reduced. For classifi-
cation, the use of the backpropagation neural network has been employed, while the 
face/non-face database has been processed using the Gabor filter. Subsequently, the 
dimension of the resulting arrays is reduced using the PCA. The BPNN is trained to 
detect facial/non-facial features using the resulting PCA. The architectural design of 
the BPNN is characterized by three strata Feed Forward Neural Network, and super-
vised learning was employed in BPNN training [28]. The function which activates the 
hidden layer is referred to as the sigmoid function. The following sub-sections provide 
brief descriptions of the three layers: 

The input layer: the input layer is the layer that receives the training patterns which 
are afterward distributed to other layers regularly. The number of input data points per 
pattern depends on the number of nodes in this layer. The number of neurons is fixed at 
(10), depending on the dimension of the input pattern vector.

The hidden layer: this layer is characterized by a specific number of neurons that 
must be synchronized to the sizes of the input and output layers and other network 

130 http://www.i-jim.org



Paper—Automated Cheating Detection based on Video Surveillance in the Examination Classes

parameters. The parameters are used for the evaluation of the system’s training and 
testing performances. 

The output layer produces the results of the neuron using a threshold function, 
where (0) denotes non-face, while (0) means face. 

Detection of face overlapping. This stage involves checking the direction of the 
face that has been detected from the image frame. In an event that the space between 
the positions of the detected faces from one frame to the other is more than a given 
threshold value, then the behavior will be judged as abnormal. The steps involved in the 
detection of face overlapping are described as follows: 

1. After the first video frame has been subjected to the process of face detection, the 
number of faces detected as well as the coordinates of each face within the frame are 
calculated and stored. 

2. In the next step, all the relevant head movements within the sequencing frames are 
detected according to a certain threshold value. The threshold value set in this paper 
is T=64. The equation below describes this process. 

 ((store(xcrdt1) + T < vctr(xcrdt1))||((store(xcrdt2)–T) > vctr(xcrdt 2)) (8)

Where (store) is made up of the number of faces that have been detected as well 
as the coordinates of each face contained in the first frame. Meanwhile, vctr is made 
up of the coordinates of the face found in the current frame as well as its corre-
sponding details. (xcrdt1, xcrdt2) – represent the x coordinates of sequence frames.

3. When the condition in equation (8) is fulfilled, it implies the presence of a significant 
head, pointing to the existence of anomaly behavior within the frame. 

4.1 Iris movement detection 

The path in which the student’s iris is moving towards to cheat using a written text 
is determined through this technique. If the system detects that the iris of the student is 
looking within the right near peripheral vision and the left near peripheral vision with 
the heads stable at one position, while the transitory period of the focus of the eyeball 
is 1, 2, and 3 seconds, then the system classifies the behavior as abnormal. There are 
four key subsections involved in the process of Iris Movement Detection: detection 
and tracking of the face, detection of facial features’ points, detection and tracking 
of regions of the eye, as well as detection of regions of the iris. The process of Face 
detection and tracking has initially been completed at the initial phase described above. 
The detection of the Facial features points is dependent on the face position that has 
been detected as well as its area; the active shape model is used in detecting the facial 
feature points. After the detection of the facial features points, the regions of the left 
and right eye which are rectangular from the face are automatically estimated. Through 
the use of a Vector Field of Image Gradient (VFOIG), the center of the iris is detected. 
This method is regarded as an efficient method because it provides very relevant details 
about circular objects like the iris. The mechanism through which the VFOID detects 

iJIM ‒ Vol. 16, No. 08, 2022 131



Paper—Automated Cheating Detection based on Video Surveillance in the Examination Classes

the middle of the circular is described in Figure 3 below. The use of VFOIG has been 
employed in this study given the circular nature of the iris since it works with circular 
objects. In this method, the orientation of every gradient vector is used in drawing a line 
across the whole image. Subsequently, the accumulator bin is increased whenever any 
of such lines pass through it. Where the majority of the lines intersect, the center of the 
iris is projected as the accumulator bin.

Fig. 3. A summary vision of iris sclera: (a) the gradient vector and movement with the 
identical alignment; (b) the gradient vector and movement with diverse alignment [29].

The VFOID employed in this work is described as follows: where Igi denotes the 
gradient vector in location Ixi, and Ic represents the possible iris center. The normal-
ized displacement vector Idi must have identical absolute orientations like the gradient 
Igi. To determine the iris’ optical center Ic

* (which is the darker part of the eye), the dot 
product of Igi and Idi should be calculated, and the global maximum of the dot product 
above the eye should be derived as represented in the equation [16] below:

 I I Idi Igic c
i

N
Targmax

N
* *� � �

�
�
�

��

�
�
�

���
�

1

1

2  (9)

Where 

 Idi
Ixi I
Ixi I

c

c

�
�

�
2

 (10)

I = 1, …, N, the displacement vector Idi should be scaled to unit length so that an 
equal weight can be achieved for all pixel positions in the image. 

4.2 Detection of hand movement

This method is specifically considered to enable the identification of the abnormal 
behavior that is common among the students, such as the exchange of incriminating 
materials. This algorithm is made up of two functions. The first function involves the 
construction of squares around the objects which are the hands, and then the algorithm 
analysis if the boundaries of the box of two examinees are connected. The second func-
tion involves checking if two students are in direct contact with each or not. There are 
four sub-stages which the hand movement detection algorithm is made up: skin detec-
tion, face detection, edges detection, and movement detection. The detection of faces 
has been completed at the first phase described above. As a first step in the process of 

132 http://www.i-jim.org



Paper—Automated Cheating Detection based on Video Surveillance in the Examination Classes

skin detection, the RGB color space is converted into a YCbCr color space. The next 
step is the extraction of skin regions based on the frame that corresponds with the 
YCbCr color space. To find the skin regions, the equations are used: 

 Cb Cb Cr Cr�� �� �� ��77 127 133 173& & & & & &  (11)

If the conditions above are satisfied, the regions of the skin color can e extracted 
from the image structure. For the human hand to be detected, the result of the detected 
face must be extracted. 

Movement detection involves identifying movement through the subtraction of 
two successive frames, and then the amount of movement is produced by the resultant 
image. If the various images are combined with equation (11), the region of dynamic 
skin color that has some activities is conserved, whereas, the areas of static skin color 
within the image are eliminated. Apart from the faces objects, the movement that occurs 
within the frames is represented by the areas of dynamic skin color.

Edges detection this process of detection aims to build squares around the hands 
that have been detected. In this work, the use of canny edge detection technique has 
been employed in detecting the edge of the hand. After the edges have been detected, 
the centroids of the hands are calculated while the squares surrounding the objects 
are pinched. This edges detection attempts to map out the boundaries of the hands 
alongside their corresponding faces [30]. Direct contact among two hand objects can 
be detected through two steps: From the first frame, a given label/flag is used to differ-
entiate and specify two hands belonging to a face. Also, the coordinates of each hand 
object are stored. Anytime a contact is detected between two hands, the system will 
need to ascertain the correspondence between the hands and the face. 2) if the system 
understands that the two hands are for two different persons, then it will raise an alarm. 
The following algorithm can be used to calculate the detection edges. 

For every contact hand areas
 If flag (handcrdt1)! = flag (handcrdt2)
  Raise alert
 End
End

Flag: the flag which is used in determining which hand belongs to which face is 
returned. 

handcrdt1, handcrdt2: denotes to the hand coordinates of the detected hand area. 

Rule-based	 performance	 classification. After the movement of the head and 
movement of the iris have been determined, the behavior is then classified by the 
proposed system. This stage involves the recognition of behavior that can be placed 
under the classify of normal or abnormal. If the system classifies the behavior of the 
student as abnormal, then more attention should be given to that student. 

iJIM ‒ Vol. 16, No. 08, 2022 133



Paper—Automated Cheating Detection based on Video Surveillance in the Examination Classes

5 Results and analysis

This sub-section presents the results that were obtained after inputting the videos 
into the system. When the output video is played, it can be observed that the irises, 
hands, and faces for all students in the input video are surrounded by a green square 
which is the frame of the input video. If abnormal behavior is detected by the system, 
the system puts a red square around the hands, irises, and face of students that are 
involved in the abnormal behavior that has been detected. The red square serves as an 
alert to the authority who is viewing the output video from the system. Figure 4 below 
shows the results. The input video’s outcome showing the usual behavior that has been 
detected is presented in Figure 4a. In this video, the squares are in green color because 
the detected behavior is normal. On the other hand, Figure 4b indicates the detection 
of abnormal behavior, whereby, a student has captured copying answers from a mobile 
phone. Similarly, the video in Figure 4c indicates the detection of abnormal behavior, 
whereby, a student is captured while taking a peek into another student’s answer sheet. 
The system surrounds the face of the student who is captured taking a peek with a red 
square so as differentiate him from his neighbor. Figure 4d presents the input video 
results, showing the exchange of cheat materials between two students. The system sur-
rounds the hands of both students with red square because their activity is recognized as 
abnormal behavior. However, each student’s face is surrounded by a green square since 
their face remains within the set threshold. 

Fig. 4. Results (a) Actual performance (b) cheat the solution from the cell-phone 
(c) one student glancing at the answer sheet of the other (d) a cheating paper is exchanged 

Between two classmates; a cheating paper is exchanged.

134 http://www.i-jim.org



Paper—Automated Cheating Detection based on Video Surveillance in the Examination Classes

6 Conclusion

In this article, an automated surveillance system that detects any abnormal behav-
ior exhibited by a student during exams has been presented. The application of this 
system is critical in detecting and recognizing an abnormal behavior (cheating) during 
an examination. This system is more efficient than humans because human error can 
emerge as a result of fatigue or sickness; these factors hamper on the performance 
of human invigilators. This paper proposes different algorithms for the identification 
and detection of the movements of the hands, eyes, and heads or students. The system 
performs the classification of the student’s activity as abnormal or normal. More so, 
the system surrounds the eyes, hands, and faces that have been identified with a green 
square, but when these body parts go beyond the set threshold, the color of the square 
changes to red, indicating an abnormal behavior. 

7 References

 [1] A. Da Veiga, L. V. Astakhova, A. Botha, and M. Herselman, “Defining organizational infor-
mation security culture—Perspectives from academia and industry,” Computers Security, 
vol. 92, p. 101713, 2020. https://doi.org/10.1016/j.cose.2020.101713

 [2] H. Tauma, and H. Alrikabi, “Enhanced data security of communication system using com-
bined encryption and steganography,” International Journal of Interactive Mobile Technol-
ogies, vol. 15, no. 16, pp. 144–157, 2021. https://doi.org/10.3991/ijim.v15i16.24557

 [3] T. Ko, “A survey on behaviour analysis in video surveillance applications,” in Video Surveil-
lance: InTech, 2011, pp. 279–294. https://doi.org/10.5772/15302

 [4] H. T. Salim, and I. A. Aljazaery, “Encryption of color image based on DNA strand and 
exponential factor,” International Journal of Online and Biomedical Engineering (iJOE), 
vol. 18, no. 3, pp. 101–113, 2022. https://doi.org/10.3991/ijoe.v18i03.28021

 [5] S.-C. Huang, “An advanced motion detection algorithm with video quality analysis for 
video surveillance systems,” IEEE Transactions on Circuits Systems for Video Technology, 
vol. 21, no. 1, pp. 1–14, 2010. https://doi.org/10.1109/TCSVT.2010.2087812

 [6] H. ALRikabi, and N. Alseelawi, “A novel method of multimodal medical image fusion 
based on hybrid approach of NSCT and DTCWT,” International Journal of Online and 
Biomedical Engineering, vol. 18, no. 3, pp. 114–133, 2022. https://doi.org/10.3991/ijoe.
v18i03.28011

 [7] G. Goo, K. Lee, H. Ka, B. Kim, W. Kim, J. Yoon, and J. Kim, “Analysis of crowded scenes 
in surveillance videos,” Can. J. Image Process. Comput. Vision, vol. 1, no. 1, pp. 52–75, 
2010. 

 [8] R. M. Alairaji, and I. A. Aljazaery, “Abnormal behavior detection of students in 
the examination hall from surveillance videos,” in Advanced Computational Para-
digms and Hybrid Intelligent Computing: Springer, 2022, pp. 113–125. https://doi.
org/10.1007/978-981-16-4369-9_12

 [9] R. A. Azeez, M. K. Abdul-Hussein, and M. S. Mahdi, “Design a system for an approved 
video copyright over cloud based on biometric iris and random walk generator using water-
mark technique,” Periodicals of Engineering Natural Sciences, vol. 10, no. 1, pp. 178–187, 
2022. 

iJIM ‒ Vol. 16, No. 08, 2022 135

https://doi.org/10.1016/j.cose.2020.101713
https://doi.org/10.3991/ijim.v15i16.24557
https://doi.org/10.5772/15302
https://doi.org/10.3991/ijoe.v18i03.28021
https://doi.org/10.1109/TCSVT.2010.2087812
https://doi.org/10.3991/ijoe.v18i03.28011
https://doi.org/10.3991/ijoe.v18i03.28011
https://doi.org/10.1007/978-981-16-4369-9_12
https://doi.org/10.1007/978-981-16-4369-9_12


Paper—Automated Cheating Detection based on Video Surveillance in the Examination Classes

 [10] D. Al-Malah, and Haider TH., “The interactive role using the Mozabook digital education 
application and its effect on enhancing the performance of eLearning,” International Jour-
nal of Emerging Technologies in Learning (iJET), vol. 15, no. 20, pp. 21–41, 2020. https://
doi.org/10.3991/ijet.v15i20.17101

 [11] L. F. Jawad, and B. H. Majeed, “The impact of teaching by using STEM approach in the 
development of creative thinking and mathematical achievement among the students of the 
fourth scientific class,” International Journal of Interactive Mobile Technologies, vol. 15, 
no. 13, pp. 172–188, 2021. https://doi.org/10.3991/ijim.v15i13.24185

 [12] D. Al-Malah, and H. ALRikabi, “Enhancement of educational services by using the internet 
of things applications for talent and intelligent schools,” Periodicals of Engineering and 
Natural Sciences (PEN), vol. 8, no. 4, pp. 2358–2366, 2020. 

 [13] P. W. Schuhmann, R. T. Burrus, P. D. Barber, J. E. Graham, and M. Elikai, “Using the 
scenario method to analyze cheating behaviors,” Journal of Academic Ethics, vol. 11, no. 1, 
pp. 17–33, 2013. https://doi.org/10.1007/s10805-012-9173-4

 [14] R. S. Rakibe, and B. D. Patil, “Background subtraction algorithm based human motion 
detection,” International Journal of Scientific Research Publications, vol. 3, no. 5, pp. 2250–
3153, 2013. 

 [15] G. Bo, S. Kefeng, Q. Daoyin, and Z. Hongtao, “Moving object detection based on improved 
ViBe algorithm,” International Journal of Smart Home, vol. 9, no. 12, pp. 225–232, 2015. 
https://doi.org/10.14257/ijsh.2015.9.12.23

 [16] W. Biao, and Z. Lin, “Improvements on Vibe algorithm for detecting foreground objects,” 
in 2016 5th International Conference on Computer Science and Network Technology 
(ICCSNT), 2016, pp. 40–44: IEEE. https://doi.org/10.1109/ICCSNT.2016.8069375

 [17] I. A. Aljazaery, A. A. Ali, and H. M. Abdulridha, “Classification of electroencephalograph 
(EEG) signals using quantum neural network,” Signal Processing: An International Journal, 
vol. 4, no. 6, p. 329, 2011. 

 [18] N. H. Barnouti, S. S. M. Al-Dabbagh, W. E. Matti, and M. A. S. Naser, “Face detection and 
recognition using Viola-Jones with PCA-LDA and square euclidean distance,” International 
Journal of Advanced Computer Science Applications, vol. 7, no. 5, pp. 371–377, 2016. 
https://doi.org/10.14569/IJACSA.2016.070550

 [19] R. S. Khairy, and A. Hussein, “The detection of counterfeit banknotes using ensemble learn-
ing techniques of AdaBoost and voting,” International Journal of Intelligent Engineering 
and Systems, vol. 14, no. 1, pp. 326–339, 2021. https://doi.org/10.22266/ijies2021.0228.31

 [20] A. S. Hussein, R. S. Khairy, and S. M. M. Najeeb, “Credit card fraud detection using fuzzy 
rough nearest neighbor and sequential minimal optimization with logistic regression,” Inter-
national Journal of Interactive Mobile Technologies, vol. 15, no. 5, pp. 24–42, 2021. https://
doi.org/10.3991/ijim.v15i05.17173

 [21] N. T. Deshpande, and S. Ravishankar, “Face detection and recognition using Viola-Jones 
algorithm and fusion of PCA and ANN,” Advances in Computational Sciences Technology, 
vol. 10, no. 5, pp. 1173–1189, 2017. 

 [22] A. Al-zubidi, R. K. Hasoun, and S. Hashim, “Mobile application to detect covid-19 pan-
demic by using classification techniques: Proposed system,” International Journal of Inter-
active Mobile Technologies, vol. 15, no. 16, pp. 34–51, 2021. https://doi.org/10.3991/ijim.
v15i16.24195

 [23] Y. C. See, and N. M. Noor, “Integrating complete Gabor filter to the random forest classifi-
cation algorithm for face recognition,” Journal of Engineering Science Technology, vol. 14, 
no. 2, pp. 859–874, 2019. 

136 http://www.i-jim.org

https://doi.org/10.3991/ijet.v15i20.17101
https://doi.org/10.3991/ijet.v15i20.17101
https://doi.org/10.3991/ijim.v15i13.24185
https://doi.org/10.1007/s10805-012-9173-4
https://doi.org/10.14257/ijsh.2015.9.12.23
https://doi.org/10.1109/ICCSNT.2016.8069375
https://doi.org/10.14569/IJACSA.2016.070550
https://doi.org/10.22266/ijies2021.0228.31
https://doi.org/10.3991/ijim.v15i05.17173
https://doi.org/10.3991/ijim.v15i05.17173
https://doi.org/10.3991/ijim.v15i16.24195
https://doi.org/10.3991/ijim.v15i16.24195


Paper—Automated Cheating Detection based on Video Surveillance in the Examination Classes

 [24] R. M. Al_airaji, I. A. Aljazaery, and S. K. Al_dulaimi, “Generation of high dynamic range 
for enhancing the panorama environment,” Bulletin of Electrical Engineering and Informat-
ics, vol. 10, no. 1, pp. 138–147, 2021. https://doi.org/10.11591/eei.v10i1.2362

 [25] B. Gopika, K. Srilaxmi, D. Alekhya, B. Rao, and B. Mohan, “Face recogition by using 
Gabor feature extraction and neural networks,” IOSR Journal of Electronics Communication 
Engineering, pp. 2278–2834, 2015. 

 [26] M. S. Da’san, “Multi-Stage Face Detection System Using Viola & Jones Method and Neural 
Networks Based on Reduced Features from Gabor Wavelet Filters,” Thesis submitted to 
Yarmouk University, Jordan, 2013. 

 [27] J. Shlens, “A tutorial on principal component analysis,” arXiv preprint arXiv:.1404.1100, 
2014. 

 [28] K. Bai, Q. An, L. Liu, and Y. Yi, “A training-efficient hybrid-structured deep neural network 
with reconfigurable memristive synapses,” IEEE Transactions on Very Large Scale Integra-
tion Systems, vol. 28, no. 1, pp. 62–75, 2019. https://doi.org/10.1109/TVLSI.2019.2942267

 [29] D. Das, M. G. Rashed, M. R. Bashar, and Y. Kuno, “Real-time iris center detection and 
tracking from video images,” in International Conference on Machine Learning and Data 
Engineering (iCMLDE2017), 2017, pp. 35–41. 

 [30] S. MohaimenianPour and R. Vaughan, “Hands and faces, fast: mono-camera user detection 
robust enough to directly control a UAV in flight,” in 2018 IEEE/RSJ International Con-
ference on Intelligent Robots and Systems (IROS), 2018, pp. 5224–5231: IEEE. https://doi.
org/10.1109/IROS.2018.8593709

8 Authors

Roa’a M. Al_airaji received the Bachelor degree from Department of Computer, 
College of Science, University of Babylon, in 2012. Received the Master degree from 
Department of software, College of Information Technology, University of Babylon, in 
2018. Currently work as assistant teacher in College of Science, University of Babylon.

Ibtisam A. Aljazaery is presently Asst. Prof. and on the faculty of Electrical Engi-
neering Department, College of Engineering, University of Babylon. Babylon, Iraq. 
Email: ibtisamalasady@gmail.com. The number of articles in national databases – 10, 
The number of articles in international databases – 5.

Haider Th. Salim ALRikabi is presently Asst. Prof. and one of the Faculty Col-
lege of Engineering, Electrical Engineering Department, Wasit University in Al Kut, 
Wasit, Iraq. He received his B.Sc. degree in Electrical Engineering in 2006 from the Al 
Mustansiriyah University in Baghdad, Iraq. His M.Sc. degree in Electrical Engineering 
focusing on Communications Systems from California State University / Fullerton /
USA in 2014. His current research interests include Communications systems with the 
mobile generation, Control systems, intelligent technologies, smart cities, and the Inter-
net of Things (IoT). Al Kut City-Hay ALRabee, Wasit, Iraq. Email: hdhiyab@uowasit.
edu.iq. The number of articles in national databases – 15, The number of articles in 
international databases – 45.

Abdul Hadi M. Alaidi is a Asst. Prof. in the Engineering College, at the Wasit 
University, Iraq. His area of research focuses on algorithm and image processing.

Article submitted 2022-02-12. Resubmitted 2022-03-10. Final acceptance 2022-03-11. Final version 
published as submitted by the authors.

iJIM ‒ Vol. 16, No. 08, 2022 137

https://doi.org/10.11591/eei.v10i1.2362
https://doi.org/10.1109/TVLSI.2019.2942267
https://doi.org/10.1109/IROS.2018.8593709
https://doi.org/10.1109/IROS.2018.8593709
mailto:ibtisamalasady@gmail.com
mailto:hdhiyab@uowasit.edu.iq
mailto:hdhiyab@uowasit.edu.iq