0% found this document useful (0 votes)
16 views29 pages

K Bank

The project report presents a 'Smart Online Proctoring System' designed to enhance the integrity of online examinations using advanced AI technologies such as facial recognition and eye-tracking. It aims to address challenges in remote assessments by providing real-time monitoring and alerts for suspicious behavior, while ensuring scalability and user-friendliness. The report outlines the project's objectives, methodology, and key features, emphasizing the importance of maintaining academic standards in digital learning environments.

Uploaded by

j2nxjwx4yg
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
16 views29 pages

K Bank

The project report presents a 'Smart Online Proctoring System' designed to enhance the integrity of online examinations using advanced AI technologies such as facial recognition and eye-tracking. It aims to address challenges in remote assessments by providing real-time monitoring and alerts for suspicious behavior, while ensuring scalability and user-friendliness. The report outlines the project's objectives, methodology, and key features, emphasizing the importance of maintaining academic standards in digital learning environments.

Uploaded by

j2nxjwx4yg
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 29

VISVESVARAYA TECHNOLOGICAL UNIVERSITY

Jnana Sangama, Belgaum-590018

A PROJECT REPORT
ON
” Smart Online Proctoring System”
Submitted in partial fulfillment of the requirements for the award of the degree of
BACHELOR OF ENGINEERING
IN
INFORMATION SCIENCE AND ENGINEERING
Submitted by

Akash Singh 1CR21IS011


Sudiksha Biradar 1CR21IS161

Under the Guidance of


Jayashree M
Assistant Professor, Department of ISE, CMRIT

DEPARTMENT OF INFORMATION SCIENCE AND ENGINEERING


CMR INSTITUTE OF TECHNOLOGY
AECS LAYOUT, ITPL PARK ROAD, BENGALURU - 560037
2024-25
CMR Institute of Technology
AECS Layout, Bengaluru-560037
Department of Information Science and Engineering

CERTIFICATE
Certified that the project work entitled ”Smart Online Proctoring System”
is a bonafide work carried out by

Akash Singh 1CR21IS011


Sudiksha Biradar 1CR21IS161
in partial fulfillment for the award of Bachelor of Engineering in Information Science
and Engineering of the Visvesvaraya Technological University, Belgaum during the year
2024-2025. It is certified that all corrections/suggestions indicated for internal
assessment have been incorporated in the report deposited in the department library.
The project report has been approved as it has satisfied the academic requirements with
respect the project work prescribed for the said Degree.

Assistant Prof. Jayashree M Dr. Jagadishwari V


Department of ISE Head of the Department
Project Guide Department of ISE

Viva

Name of the Examiner Signature with date

1.

2.
ABSTRACT

A Smart online proctoring system is an advanced solution designed to uphold the


integrity of online examinations by leveraging cutting-edge AI technologies. It
incorporates features such as facial recognition to verify student identity, eye-
tracking to monitor focus, and screen monitoring to ensure compliance with
exam rules. These systems are capable of detecting suspicious behaviors, such as
unusual movements, the presence of unauthorized individuals, or attempts to
access prohibited resources. Real-time monitoring allows for immediate
intervention in case of rule violations, while post-exam analysis provides detailed
reports for educators to review. Additionally, the system can track student
activity patterns, offering insights into engagement and potential areas of
concern. By addressing the growing challenges of remote education, smart
proctoring systems promote fairness and reliability in assessments, ensuring that
academic standards are maintained in a digital learning environment. These
systems also enhance scalability and convenience, enabling institutions to
conduct secure exams for large numbers of students regardless of location.
ACKNOWLEDGMENT

The satisfaction and euphoria that accompany a successful completion of any task
would be incomplete without the mention of people who made it possible. Success
is the epitome of hard work and perseverance, but steadfast of all is encouraging
guidance.

So, it is with gratitude that I acknowledge all those whose guidance and
encouragement served as beacon of light and crowned our effort with success. I
would like to thank Dr. Sanjay Jain, Principal, CMRIT, Bangalore, for providing an
excellent academic environment in the college and his never-ending support for
the B.E program.

I would like to express my gratitude towards Dr. Jagadishwari V, Professor


and HOD, Department of Information Science and Engineering CMRIT,
Bangalore, who provided guidance and gave valuable suggestions regarding the
project.

I consider it a privilege and honor to express my sincere gratitude to our internal


guide Assistant Prof. Jayashree M, Department of Information Science and
Engineering, CMRIT, Bangalore, for their valuable guidance throughout the tenure
of this project work.

I would like to thank all the faculty members who have always been very
cooperative and generous. Conclusively, I also thank all the nonteaching staff
and all others who have done immense help directly or indirectly during our
project.

Akash Singh
Sudiksha Biradar

ii
Contents
Abstract i

Acknowledgment ii

Table of Content iii

List of Figures v

1 Introduction 1
1.1 Problem Statement.......................................................................................... 1
1.2 Objectives ....................................................................................................... 1
1.3 Scope of Project .............................................................................................. 2
1.3.1 Key Features and Functionality ............................................................... 2
1.3.2 Technology Stack .................................................................................. 2
1.3.3 Target Audience ................................................................................... 2
1.3.4 Functional Scope .................................................................................. 2
1.3.5 Non- Functional Scope .......................................................................... 3
1.3.6 Challenges to consider ....................................................................... 3
1.3.7 Future enhancements ........................................................................ 3
1.4 Proposed Methodology .................................................................................. 3
1.4.1 Requirement Analysis ........................................................................ 4
1.4.2 System Design .................................................................................... 4
1.4.3 Technology Selection .......................................................................... 5
1.4.4 Implementation ................................................................................. 5
1.4.5 Testing and validation ........................................................................ 5
1.4.6 Deployment ....................................................................................... 6

2 Literature Survey 7

3 System Design 9
3.1 Data Flow Diagram ........................................................................................ 10
3.2 System Architecture ...................................................................................... 11

4 Implementation 12
4.1 Implementation Details ................................................................................ 12
4.2 Algorithm Use ............................................................................................... 12
4.3 Experimental Setup ...................................................................................... 12
4.4 Algorithm Used for Comparison .................................................................. 13
4.5 Performance Metric ..................................................................................... 13
4.6 Why These Metrics and Algorithms were chosen ....................................... 13

iii
5 Results 14
5.1 Performance Metrics .................................................................................... 14
5.2 Comparative Analysis .................................................................................... 14
5.3 Model Analysis ............................................................................................. 15
5.4 Short Comings .............................................................................................. 15
5.5 Future Improvements .................................................................................. 16
5.6 Output Screenshots ...................................................................................... 16

6 Testing 17
6.1 Test Case Development ................................................................................. 17
6.2 Test Phases .................................................................................................... 18
6.3 User Feedback ............................................................................................... 18

7 Conclusion 20
7.1 Contributions and Utility ................................................................................. 20
7.2 Scope for Further Work ................................................................................... 20

8 References 22

iv
List of Figures
1 3.1 Data Flow Diagram .................................................................................. 10
2 3.2 System Architecture ................................................................................ 11
3 5.1 Accuracy Comparison ............................................................................. 14
4 5.2 ROC Curve ............................................................................................... 15
5 5.3 Output 1 ................................................................................................. 16
6 5.4 Output 2 ................................................................................................. 16

v
Smart Online Proctoring System

1 INTRODUCTION
Global acceptance of remote work practices and the quick growth of online education are
the main drivers of the rising need for remote assessments. Finding safe and trustworthy
ways to administer tests has grown more important as corporations and educational
institutions move to digital platforms. Though they offer accessibility and flexibility,
remote assessments also present serious problems for preserving exam integrity.
Maintaining academic and professional standards requires making sure these tests are fair
and guarding against cheating with no human oversight. The significance of creating
strong, technologically advanced solutions to protect the legitimacy of remote assessments
while providing administrators and participants with a smooth, scalable experience is
underscored by this.

1.1 Problem Statement


The difficulties in maintaining exam integrity are highlighted by the increasing demand for
distant assessments in online learning and international employment. Manual proctoring is
expensive, time-consuming, and prone to mistakes, and simple virtual systems don't have
sophisticated AI, which results in false positives and insufficient detection of cheating. This
leaves a gap in terms of guaranteeing efficiency, scalability, and justice. With real-time
alarms, environment scanning, and cloud-based scalability, the project suggests an AI-
powered virtual proctoring system that will improve exam integrity, lower expenses, and
provide a smooth, scalable solution for international remote assessments.
1.2 Objectives
• Create a real-time artificial intelligence algorithm to identify suspicious activity during
online tests, such as avoiding eye contact, having numerous people present, or using
unapproved equipment.
• Use environment-scanning technologies to automatically identify resources that are
restricted, like printed materials, mobile devices, and secondary screens.
• Create a cloud-based system that can keep an eye on thousands of students at once,
enabling educational institutions to expand their use of online tests without
compromising the integrity of the results.
• By fine-tuning AI systems to distinguish between typical and questionable behaviour,
we can reduce false positives and improve user experience.

Dept. of ISE, CMRIT, Bengaluru 1


Smart Online Proctoring System

1.3 Scope of Project


1.3.1. Key Features and Functionalities
• Face Detection and Recognition: To verify the identity of the student and prevent
impersonation.
• Eye Movement Tracking: To detect if the student is looking away from the screen,
which may indicate cheating.
• Head Movement Detection: Detecting unusual head movements or distractions
during the exam.
• Environment Monitoring: Using cameras to monitor the surrounding environment
for any potential cheating devices.
• Real-Time Alerts: The system will alert proctors if any suspicious activity is
detected.
• Recording the Session: Storing recorded video/audio for review by the instructor or
administrator.
• AI and Machine Learning Models: The system will utilize machine learning
algorithms for analysis, pattern recognition, and detecting abnormalities in student.
1.3.2. Technology Stack
The system is built using the following technologies:
• Python for programming.
• OpenCV for image processing and video analysis.
• TensorFlow for machine learning models related to facial recognition and analysis.
• Webcam for video surveillance.
• Face Recognition Libraries such as dlib.
1.3.3. Target Audience
• Educational institutions that offer online courses.
• Examination authorities who want to ensure the integrity of online assessments.
• E-learning platforms seeking to improve the security and fairness of their testing
systems.
1.3.4. Functional Scope
• Proctoring Setup: Setting up the proctoring system for an exam, including the
configuration of cameras and environment monitoring.
• Exam Session: Students take exams while the system continuously monitors their .

Dept. of ISE, CMRIT, Bengaluru 2


Smart Online Proctoring System

• Suspicious Activity Detection: Real-time alerts if suspicious behaviour is detected


(e.g., looking away, presence of multiple faces).
• End of Exam: After the exam, the recorded session is reviewed, and reports are
generated for exam administrators.
1.3.5. Non-Functional Scope
• Scalability: The system should be able to handle a large number of simultaneous
exam sessions without lag.
• Security: The system must ensure that no personal data is exposed or misused. Data
encryption and secure storage are essential.
• User-Friendly Interface: The system interface should be simple and intuitive for
both students and exam proctors.
• Accuracy: The detection algorithms must have high accuracy in detecting cheating
behaviours, ensuring minimal false positives.
1.3.6. Challenges to Consider
• Privacy Concerns: Since the system uses webcams and tracks facial expressions and
movements, ensuring user privacy will be a challenge.
• False Positives/Negatives: Ensuring the system does not wrongly flag normal
behaviours as suspicious or miss actual suspicious actions.
• Adaptability to Different Environments: Ensuring that the system works effectively
in various lighting conditions, camera qualities, and environments.
• Real-time Processing: Handling real-time analysis of video feeds while maintaining
system performance.
1.3.7. Future Enhancements
• AI Training: Continuously training the AI models with new data to improve
detection accuracy.
• Behavioural Analytics: Extending the system to analyze more behavioral patterns
(e.g., facial expressions, emotional responses).
• Mobile App Support: Expanding the system to support mobile devices for remote
exam-taking.
This scope outlines a high-level view of the project and provides a comprehensive
approach to developing and deploying an intelligent online exam proctoring system,
focusing on fairness, security, and user experience.

Dept. of ISE, CMRIT, Bengaluru 3


Smart Online Proctoring System

1.4 Proposed Methodology

1.4.1. Requirement Analysis

• Objective: Understand the specific needs of the system, such as the ability to detect
cheating, verify identity, and monitor exam behaviour.
• Data Collection: Gather requirements from educational institutions, exam
administrators, and students. This includes determining necessary features (e.g.,
facial recognition, eye movement tracking, real-time alerts).
• System Constraints: Consider system performance, scalability, and security
requirements (e.g., encryption for privacy).
Deliverable: A clear list of requirements, including functional and non-functional
specifications.

1.4.2. System Design

• Architecture Design:
o Frontend: The user interface for both the students and exam administrators
(e.g., dashboards, video feed view, alerts).
o Backend: Servers to handle video streaming, real-time processing, and
storage of exam data.
o Video Surveillance System: A setup of webcam or camera systems that
capture exam sessions for analysis.
• Database Design: Plan for storing user profiles, exam data, behaviour logs, and
video recordings. Ensure secure data storage practices (e.g., encryption).
• Detection Mechanism Design:
o Face Recognition: Implement algorithms to detect and verify the student’s
face, using libraries such as dlib.
o Behavioural Detection: Design algorithms to monitor and track eye
movements, head movements, and other behaviours (e.g., detecting if a
student is looking away from the screen).
o Suspicious Activity Detection: Define behaviours such as looking away,
excessive blinking, or external devices that could trigger alerts.

Dept. of ISE, CMRIT, Bengaluru 4


Smart Online Proctoring System

1.4.3. Technology Selection

• Machine Learning Framework:


o TensorFlow: For training models related to facial recognition and abnormal
behaviour detection.
o OpenCV: For real-time image processing (e.g., face and object detection).
• Programming Language: Python for backend logic and AI model integration.
• Web Framework: Django or Flask for creating the web interface and managing the
communication between frontend and backend.
• Camera and Video Tools: Use webcams or other cameras, ensuring the system can
process live video feeds in real-time.

1.4.4. Implementation

• Face Detection and Recognition: Implement a face detection model to verify the
student's identity at the start of the exam. This ensures that the right individual is
taking the test. Use a pretrained model like Haar Cascades, dlib, or FaceNet for real-
time facial recognition.
• Behavioral Monitoring:
o Eye Tracking: Use algorithms to track eye movements and detect if the
student looks away from the screen, which may indicate cheating.
o Head Movement Detection: Track the student's head movement to detect
unusual behavior (e.g., looking off-screen frequently).
• Suspicious Activity Detection: Train the system to detect suspicious behavior, such
as the presence of multiple faces, external devices (e.g., phones), or interactions
with objects that could signal cheating.
• Alert System: When suspicious behavior is detected, an alert is sent to the exam
administrator or proctor, who can take action.
• Real-time Video Processing: Process video streams in real-time using OpenCV to
detect anomalies during the exam. This includes continuous monitoring for facial
expressions, head positioning, and other behavioral indicators.

1.4.5. Testing & Validation

• Unit Testing: Test individual components such as facial recognition, eye tracking,
and head movement detection for correctness and accuracy.

Dept. of ISE, CMRIT, Bengaluru 5


Smart Online Proctoring System

• Integration Testing: Test how the different components (e.g., detection algorithms,
video processing, and alert systems) work together.
• System Testing: Test the system under different conditions, such as varying
lighting, camera quality, and background noise.
• Accuracy and Performance: Test the accuracy of the face recognition and
behavioural detection systems by comparing results against a dataset of known
behaviours. Evaluate the system’s performance in terms of real-time processing and
resource consumption (e.g., CPU, memory).

1.4.6. Deployment

• Cloud Deployment: If the system is intended to be used for multiple exam sessions,
deploy it on a cloud platform (e.g., AWS, Google Cloud) for scalability.
• Web Interface Setup: Host the front-end application (student dashboard, proctor
dashboard) and the backend logic on a server.
• Security Considerations: Ensure data privacy by encrypting video feeds and student
data during storage and transmission. Implement security measures such as HTTPS,
two-factor authentication for proctors, and secure storage of sensitive data.
• User Training: Provide guides or training for students and exam administrators on
how to use the system.

Dept. of ISE, CMRIT, Bengaluru 6


Smart Online Proctoring System

2 LITERATURE SURVEY
[1] A literature review on AI-based proctoring systems for online exams is presented in the
paper "Online Exam Proctoring System Based on Artificial Intelligence," which also
discusses developments in behavior analysis, user authentication, and real-time cheating
detection. It investigates computer vision and machine learning methods that track the eye
movements, body language, and environmental changes of candidates. Concerns about
privacy and ethics are also covered in the study, emphasizing the need for balance.
protection of examinees' rights while prioritizing ongoing innovation to enhance user
experience and detection precision

[2] Online exam proctoring systems use AI-driven tools to monitor and stop cheating
during remote assessments, according to a literature review of the subject. They frequently
use techniques like audio detection, gaze tracking, and facial recognition to guarantee the
integrity of the test. While some studies focus on privacy, data security, and accessibility
concerns, others examine the efficacy of various proctoring methods. The ethical issues of
managing personal data and ensuring inclusion in proctoring environments, as well as
striking a balance between security and user experience, are the main topics of recent
research.

[3] The study by Adiy Tweissi, Wael Al Tawie, and Dalia Al Eisawi compares the efficacy
of traditional human proctoring with AI-based automatic proctoring in online tests. It
discovers that by employing techniques like LSTM networks for behavior analysis, AI is
able to recognize dishonest behaviors like speech, unauthorized screen activity, and
suspicious motions. However, there are drawbacks, such as sporadic false positives and
biases in facial recognition accuracy brought on by changes in skin tone and lighting.
Accordingly, AI proctoring is beneficial but has to be improved for maximum accuracy.

[4] In order to detect cheating, Vidhya SG and colleagues' "AI-Based Proctoring System
for Online Tests" describes a real-time exam proctoring solution that makes use of AI
capabilities including voice analysis, facial recognition, and CNNs. The system tracks
possible misconduct, identifies unauthorized goods, and keeps an eye on pupils'
concentration. In order to successfully prevent exam cheating in remote locations, it offers
capabilities like candidate verification and mobile phone detection.

Dept. of ISE, CMRIT, Bengaluru 7


Smart Online Proctoring System

[5] Niharika G. N. and Dr. Sankhya N. Nayak's "Artificial Intelligence Based Online
Examination Proctoring System" offers an AI-powered proctoring solution to improve
exam integrity. CNN, Haar Cascade, and YOLO V3 are examples of computer vision and
machine learning algorithms that are used in the system to track behaviors including head
movement, gaze direction, and the detection of prohibited objects (such cell phones). It
provides a scalable and automated method of remote proctoring while lowering the
requirement for manual oversight by flagging suspicious actions for human evaluation.

Objective/ Existing
Reference Technique Advantages Demerits
Features Algorithms
AI-driven
Enhance Likely involves Improved Possible
solutions for
[1] security during machine learning monitoring privacy
maintaining
online exams for detection. capabilities issues.
exam integrity
Convolution Ethical
Facial High
Real-time Neutral concerns
recognition, accuracy in
[2] proctoring Networks regarding
voice analysis, detecting
using AI. (CNNs) for personal
and CNNs cheating.
visual analysis. data
Objective Human
Comparison
Assess Not specified; comparison proctoring
of AI-based vs
[3] effectiveness focus on of different limitations
Human
and validity comparison proctoring in
proctoring
types. scalability
Various AI
tools like Maintaining Computer
Robust and Privacy and
facial integrity and Vision, Machine
[4] scalable accessibility
recognition prevent Learning, Eye
solutions challenges.
and gaze cheating Tracking
tracking
Eye-tracking,
Real-time Balancing
Computer body language Accurate
cheating security
vision, analysis, and monitoring
[5] detection and with
behaviour environmental of exam
user privacy
analysis monitoring conditions
authentication concerns.
techniques
Table 2.1: Comparison of all previous models

Dept. of ISE, CMRIT, Bengaluru 8


Smart Online Proctoring System

3 System Design
This chapter offers a thorough explanation of the Smart Online Proctoring System's
architecture and implementation. To guarantee smooth and safe proctoring during online
tests, the model makes use of cutting-edge machine learning algorithms and computer
vision techniques. In order to host, execute, and use the application effectively, the chapter
also lists the minimal hardware and software requirements.

A processor with capabilities comparable to an Intel Core i5 or above is necessary for the
system to operate at its best, providing enough processing power to manage real-time video
and audio analysis. Multitasking and memory-intensive operations, such as processing
video feeds and executing machine learning models, require at least 8 GB of RAM. A
minimum of 256 GB SSD is needed for storage in order to support fast data access and
system responsiveness. To record real-time video and audio inputs throughout the test,
additional peripherals like an HD webcam (720p or above) and an external or built-in
microphone are necessary. Additionally, to ensure continuous communication between the
client and the provider, a reliable broadband internet connection with a minimum speed of
10 Mbps is essential.

On the software side, the program is made to be compatible with a wide range of
contemporary operating systems, such as Windows 10, macOS, or Linux. Because of its
extensive library and framework ecosystem, Python (version 3.8 or higher) is the most
widely used programming language. Important dependencies include Scikit-learn for
further data analysis activities, TensorFlow or Keras for creating machine learning models,
and OpenCV for processing images and videos. It is recommended that end users utilize
the most recent versions of web browsers, like Google Chrome or Mozilla Firefox, to access
the program for best results.

This configuration guarantees the Smart Online Proctoring System's stability, scalability,
and ability to provide dependable performance in a variety of settings.

Dept. of ISE, CMRIT, Bengaluru 9


Smart Online Proctoring System

3.1 Data Flow Diagram

Fig 3.1: Data flow diagram

Dept. of ISE, CMRIT, Bengaluru 10


Smart Online Proctoring System

3.2 System Architecture

Fig 3.2: System Architecture

Dept. of ISE, CMRIT, Bengaluru 11


Smart Online Proctoring System

4 IMPLEMENTATION
4.1 Implementation Details

Machine learning and computer vision techniques are combined in the proposed Smart
Online Proctoring System to identify and prevent instances of cheating during online
exams. Real-time video analysis for facial recognition and detection, behavior monitoring
with posture and gaze detection, and audio analysis for identifying unwanted noises are the
three main components of the implementation. By integrating these modules into a web-
based platform, examiners may keep an eye on examinees from a distance.
TensorFlow/Keras is used to create deep learning models, OpenCV is used for video
processing, and Python is used to develop the backend with frameworks like Flask for web
services.

4.2 Algorithm Use

For face detection and recognition, the system employs a Convolutional Neural Network
(CNN), using pre-trained models such as Dlib or Haar Cascades for initial face localization
and then fine-tuned CNNs for facial recognition. The system estimates the direction of the
eye using a combination of CNN-based regression models and geometric-based techniques
for gaze detection. For behavior analysis, a Recurrent Neural Network (RNN) is used to
identify suspicious activity patterns, such absenteeism or recurrent head motions.

4.3 Experimental Setup


In the controlled setting of the experiment, individuals took practice tests in a variety
settings, including varied camera settings, different lighting, and different degrees of
adherence to the exam guidelines. The system was tested on a dataset of recorded exam
sessions that comprised both regular behavior and pre-established cheating situations,
such as looking away from the screen, using a device without autohorication, and having
many people in the frame.

An Intel Core i7 processor, 16 GB of RAM, an NVIDIA GeForce GTX 1660 GPU, and a
reliable broadband internet connection enabled the local computer used for the research.
Using publically accessible datasets like GazeCapture for gaze detection and Labeled Faces
in the Wild (LFW) for face recognition, the models were trained and verified.

Dept. of ISE, CMRIT, Bengaluru 12


Smart Online Proctoring System

4.4 Algorithms Used for Comparison

A comparison was made between the suggested system's performance and:

1. Haar Cascades (OpenCV): An old-fashioned approach to facial recognition.


2. The object detection model YOLOv5 (You Only Look Once): Based on deep
learning.
3. Support Vector Machines (SVM): Used in more straightforward configurations for
behaviour classification.

These algorithms were picked for comparison since they are widely used and have different
approaches (deep learning vs. classical). The capabilities of contemporary real-time object
detection are demonstrated by YOLOv5, Haar Cascades offer a standard for classical
computer vision, and SVM provides a low-cost substitute for behavior analysis.

4.5 Performance Metrics

The following metrics were used to assess the system's and comparison algorithms'
performance:

1. Accuracy: The proportion of behaviors or actions that are correctly identified.


2. Precision and Recall: To gauge how well the system can identify illegal activity while
reducing false positives and negatives.
3. F1-Score: The trade-off-balancing harmonic mean of recall and precision.
4. Processing Speed: To assess performance in real time, processing speed is measured
in frames per second (FPS).
5. System Latency: The amount of time it takes to investigate and report questionable
activities.

4.6 Why These Metrics and Algorithms Were Chosen

An exam proctoring tool must have both accuracy and real-time effectiveness, which are
reflected in the metrics that were chosen. The algorithms were selected on the basis of their
computing efficiency and applicability to fundamental tasks such as behavior analysis and
face detection. A comparison of traditional and contemporary methods shed light on the
trade-offs between resource requirements and performance.

Dept. of ISE, CMRIT, Bengaluru 13


Smart Online Proctoring System

5 RESULTS
5.1 Performance Metrics
We examined the model using the following metrics in order to assess the system's
performance:
The percentage of outputs that are accurately identified is known as accuracy.
When comparing all positive detections, precision shows the percentage of true positive
detections (e.g., successfully recognizing cheating occurrences without false alarms).
F1 Score: When data is unbalanced, this statistic, which combines precision and recall.
By illustrating the trade-off between sensitivity (true positive rate) and specificity (false
positive rate), the Receiver Operating Characteristic (ROC) curve provides information
about threshold settings.

5.2 Comparative Analysis


To illustrate the advantages or shortcomings of the selected methodology, the outcomes
were contrasted with baseline methods such as additional machine learning models.
Accuracy Comparison in Graph 1: The suggested approach outperforms the baseline
models (e.g., Rule-Based: 75%, SVM: 85%) with an accuracy of 92%, as the graph
demonstrates. Better robustness in identifying irregularities during online tests is indicated
by this.

Fig 5.1: Accuracy Comparison

Dept. of ISE, CMRIT, Bengaluru 14


Smart Online Proctoring System

The ROC curve for the model demonstrates a high area under the curve (AUC = X.XX),

indicating strong discriminatory power in separating normal activities from suspicious .

Fig 5.2: ROC Curve

5.3 Model Analysis

92% of instances of cheating were successfully identified by the proctoring model.


However, some false positives were displayed because normal head motions were
suspicious.
In normal lighting, the face detection accuracy was 95%; in poor lighting, it decreased to
78%.

5.4 Shortcomings
1. Problems with Cloud Integration: Compatibility issues with APIs and inconsistent
deployment environments made it difficult to integrate the cloud seamlessly. By testing
across a wider variety of cloud platforms and making sure deployment
2. Environmental Restrictions: Inadequate lighting and camera quality made the system
less effective and resulted in missed detections. Performance can be enhanced by
employing picture enhancement algorithms or adaptive illumination approaches.
3. Data Restrictions: The dataset used for training lacked diversity in terms of user
demographics and environmental variations, which might have caused biases in
predictions. Expanding the dataset can make the system more robust.

Dept. of ISE, CMRIT, Bengaluru 15


Smart Online Proctoring System

5.5 Future Improvements


Adding cutting-edge computer vision methods, such as YOLO, for real-time object model.
Utilizing pre-configured cloud services, such as Google Cloud AI or AWS Lambda, to
functionality.
Adding calibration modes to the system allows customers to customize it for themselves,
which enhances performance.

5.6 Output Screenshots

Fig 5.3: Candidate has not alerts

Fig 5.4: Candidate has alert for eye and head position

Dept. of ISE, CMRIT, Bengaluru 16


Smart Online Proctoring System

6 TESTING
6.1 Test Case Development
Test
Case Test Scenario Input/Action Result Status
ID
Exam Proctoring Move out of the System detects absence
TC01 Pass
Detection camera frame and flags as suspicious
Face Recognition Test with low Detect face with
TC02 Pass
under Lighting lighting decreased accuracy
Multiple Faces in Add another person System flags multiple
TC03 Pass
Frame to the frame faces
Look off-screen for System detects off-
TC04 Gaze Detection Pass
10 seconds screen gaze and flags it
System detects and
Try any kind of lip
TC05 Mouth Movement notifies that mouth is Pass
movement
open
One person present
System recognises and
in the frame whose
TC06 Face Identification displays the persons Pass
data is stored in
name
database
Move closer and System displays the
Distance
TC07 away from the distance between the Pass
Estimation
screen person and the screen
Detects the photo and
Show ID in front of
TC08 ID card detection name and stores the data Pass
screen
into database
System
Run 20 instances Application remains
TC09 Performance under Pass
simultaneously responsive
Load

Table 6.1: Test Cases

Dept. of ISE, CMRIT, Bengaluru 17


Smart Online Proctoring System

6.2 Testing Phases


Examining Individual Units:

• Justification: Testing individual modules in isolation ensures that each component


functions correctly on its own. By conducting these tests under controlled
conditions, the system’s core functionalities—such as audio analysis, gaze
detection, and face detection—were verified to operate as expected. This step is
essential to ensure that each module is free from errors and can contribute
effectively when integrated.

Examining Integration:

• Justification: Once individual modules are tested, it's important to evaluate how
well they interact with each other. In this case, the system was found to have low
latency and smooth communication between components. This indicates that the
integration was successful, ensuring that the data flows seamlessly between the
audio, gaze, and face detection modules. Low latency is particularly crucial for real-
time systems, where delays could impact user experience.

Examining Loads:

• Justification: Load testing is crucial to understand how the system performs under
varying levels of user activity. Simulating 50 concurrent users allowed the team to
assess the system’s ability to handle multiple interactions simultaneously. The
results show that the application can support 30 users without issues, but beyond
that, minor lags were observed. This suggests that the system has a scalable
performance limit, and optimization might be needed to handle more users
simultaneously without introducing significant delays. This helps identify potential
areas for improvement in handling higher user loads.

6.3 User Feedback


• Ease of Use: 9/10 found the UI simple to use, indicating an intuitive design.
• Accuracy: 8/10 noted occasional false positives but overall correct identification
of suspicious behavior, suggesting the system is mostly accurate but could benefit
from refinement.

Dept. of ISE, CMRIT, Bengaluru 18


Smart Online Proctoring System

• Performance: 7/10 experienced performance issues on older devices, indicating


the need for optimization on lower-end hardware.
• Suggestions: Users recommended reducing false positives for normal head
motions, providing clearer instructions before exams, and improving performance
in low light. These suggestions address specific issues and aim to enhance system
reliability and user experience.

Dept. of ISE, CMRIT, Bengaluru 19


Smart Online Proctoring System

7 CONCLUSION
Using machine learning and computer vision techniques, this research created an Intelligent
Online Exam Proctoring System to improve the reliability of online tests. By combining
modules for face detection, gaze tracking, behavior monitoring, and audio analysis, the
system makes it possible to identify suspicious activity in real time, such as gaze deviation,
out-of-frame behavior, and unauthorized noise. The implementation made use of a reliable
architecture that included gaze estimation, behavior classification using RNN, and face
recognition based on CNN. In order to show the system's resilience and real-time
performance, it was tested in a variety of conditions, such as dim lighting, numerous users,
and network outages.

7.1 Contributions and Utility


1. Contributions:

• Proposed Model: created a thorough proctoring system that uses a variety


of machine learning methods to identify anomalies.
• Algorithmic optimization: enhanced gaze detection and facial recognition
components for real-time use.
• The development of an intuitive interface through user-centric design made
it easier for both examiners and examinees to use.
• Testing Framework: Extensive testing was carried out to ensure scalability
and reliability, including unit, integration, and load tests.
2. Utility:
• By offering a scalable and affordable substitute for manual invigilation, the
system lowers administrative burden.
• Ensures academic integrity by detecting and flagging suspicious behaviors
during exams.
• Simplifies examination procedures by integrating with current Learning
Management Systems (LMS).

7.2 Scope for Further Work


1. Reducing False Positives:

Dept. of ISE, CMRIT, Bengaluru 20


Smart Online Proctoring System

• Improve gaze and behavior detection algorithms to lessen the number of


times that normal motions are interpreted as suspicious.

2. Improved Lighting Adaptability:

• Using sophisticated preprocessing methods or hardware integration,


increase the reliability of face detection and gaze tracking in low light.

3. Cross-Device Compatibility:

• Make the system more accessible by optimizing it for a range of gadgets,


such as tablets and smartphones.

4. Multi-Language Support:

• Expand the platform to accommodate more languages for localizing the user
interface and audio analysis.

5. Advanced Anomaly Detection:

• To improve adaptation to erratic behaviors, apply sophisticated anomaly


detection methods utilizing reinforcement learning.

6. Integration with Biometric Verification:

• Incorporate biometric authentication techniques, such as fingerprint or iris


scanning, into the system to enhance security.

Dept. of ISE, CMRIT, Bengaluru 21


Smart Online Proctoring System

REFERENCES
[1] Mr. Vidhya SG, Ms. Hema GA, Ms. Jeevitha MG, Ms. Nischitha KB, Ms.Vandana,
"AI-BASED PROCTORING SYSTEM FOR ONLINE TESTS", International
Research Journal of Modernization in Engineering Technology and Science,
Vol.04 Issue 07/July-2022, DOI

[2] Vinothini Kasinathan, Chou Ee Yan, Aida Mustapha, Vazeerudeen Abdul Hameed,
Tham Hoong Ching, Vinesh Thiruchelvam, "ProctorEx: An Automated Online
ExamProctoring System", Mathematical Statistician and Engineering Applications
Vol. 71 No. 3s2 2022, DOI

[3] Shilpa Manohar Satre and Shankar M. Patil, "Online Exam Proctoring System Based
on Artificial Intelligence", Journal, Vol. May 2023,
10.1109/IConSCEPT57958.2023.10170577

[4] Adiy Tweissi, Wael AI Etawai and Dalia AI Eisawi,"The Accuracy of AI-Based
Automatic Proctoring in Online Exams", The Electronic Journal of e-learning, Vol.
20 issue 4 2022, DOI

[5] Niharika G N, Dr. Sankhya N Nayak, "Artificial Intelligence Based Online


Examination Proctoring System", International Journal for Research in Applied
Science & Engineering Technology (IJRASET) Vol.11 Issue IX Sep 2023, DOI

[6] Youtube (www.youtube.com)

[7] Github (www.github.com)

[8] Anaconda Distribution (www.anaconda.com)

[9] IEEE Xplore (www.ieeexplore.ieee.org)

Dept. of ISE, CMRIT, Bengaluru 22

You might also like