A

PROJECT REPORT
ON

‘DRIVE ALERT’
(A REAL TIME COMPUTER VISION APPROACH)

SUBMITTED TO THE

DEPARTMENT OF COMPUTER APPLICATION

In partial fulfillment for the award of the degree

Master of Computer Application

Under the Guidance of

Ms. Shobha G (Head of Department)

Prof. Ms. Priyanka Sharma

SUBMITTED BY

PRACHI KATIYAR (2300910140039)

PRIYANSHI TRIPATHI (23009101400043)

SESSION: 2024-25
JSS ACADEMY OF TECHNICAL EDUCATION NOIDA
DR. A.P.J ABDUL KALAM TECHNICAL UNIVERSITY, LUCKNOW
 DECLARATION

We hereby declare that the project entitled “Driver Safety System – A Real-Time Computer Vision
Approach” is a bona fide record of work carried out by us during our Master of Computer Applications
(MCA) program at JSS Academy of Technical Education, Noida. This project is submitted in partial
fulfillment of the requirements for the award of the degree and represents our original work.
The system developed includes two main modules:

1. Driver Drowsiness and Yawn Detection

2. Lane and Car Detection

These modules aim to increase road safety through real-time monitoring of driver alertness and
surroundings, using advanced computer vision techniques such as facial landmark detection and lane line
recognition.

All technologies, tools, and research used have been properly acknowledged.

Signature of Student

• Prachi Katiyar (2300910140039) Place: Noida, U.P

Signature: - Date: -

• Priyanshi Tripathi (2300910140043)

Signature: -

Internal Guide
 CERTIFICATE

Certified that the Project Report entitled “Drive Alert: A real time Computer Vision approach” is the
Bonafide work of Ms. Prachi Katiyar (2300910140039) and Ms. Priyanshi Tripathi (2300910140043),
who carried out the project work under my supervision.

Certified further that to the best of my knowledge, this work does not form part of any other project report
or dissertation submitted elsewhere for the award of any degree.

Signature of Student

• Prachi Katiyar (2300910140039) Place: Noida, U.P

Signature: - Date: -

• Priyanshi Tripathi (2300910140043)

Signature: -

Signature of Internal Guide Signature of Project in-charge

(Head of Department)
 ACKNOWLEDGEMENT
We would like to express our deep gratitude to our project guides, Ms. Shobha G (Head of
the Department) and Ms. Priyanka Sharma Assistant Professors, Department of Master
of Computer Applications, JSS Academy of Technical Education, Noida, for their
invaluable guidance, immense encouragement, and unwavering support throughout the
project.

We are extremely grateful to Ms. Shobha G, Head of the Department, Master of

Computer Applications, for providing us with the necessary facilities and resources to
complete our project work successfully. We extend our heartfelt thanks to the Principal and
Management of JSS Academy of Technical Education, Noida, for their constant
encouragement and cooperation, which enabled us to carry out this work.

Our sincere thanks go to all the teaching faculty of the Department of Master of Computer
Applications, whose constructive suggestions during reviews significantly contributed to
the successful completion of our project. We are equally thankful to the non-teaching staff
of the department for their assistance and support.

We are deeply thankful to our parents, friends, and classmates for their continuous
encouragement and moral support throughout our project journey. Lastly, we extend our
gratitude to everyone who supported us directly or indirectly in the successful completion
of this project.

1
[Link]. Contents Page No.

1 Declaration ii

2 Certificate iii

3 Acknowledgement 1

4 Abstract 2

5 List of Abbreviations 3

6 Chapter 1: Introduction 4

6.1 1.1 Introduction 4

6.2 1.2 Purpose 5

6.3 1.3 Project Objectives 5

6.4 1.4 Project Flow 6

6.5 1.5 Project Scope 6

6.6 1.6 Problem Definition 7

6.7 1.7 Module Overview 7

6.8 1.8 Impact and Application Area 8

7 Chapter 2: The Overall Description 9

7.1 2.1 Product Perspective 9

7.2 2.1.1 User Interface 9

7.3 2.1.2 Hardware Interfaces 9

7.4 2.1.3 Software Interfaces 10

7.5 2.1.4 Memory Constraints 10

7.6 2.2 Product Functions 10

7.7 2.3 User Characteristics 11

2
[Link]. Contents Page No.

7.8 2.4 Constraints 11

8 Chapter 3: Specific Requirements 12

8.1 3.1 Requirement Specification 12

8.2 3.2 Functional Requirements 12

8.3 3.3 Non-Functional Requirements 13

8.4 3.4 External Requirements 13

8.5 3.5 User Requirements 14

8.6 3.6 Hardware Requirements 14

8.7 3.7 Software Requirements 14

8.8 3.8 Software Product Features 15

8.9 3.9 Modules 16

8.10 3.10 System Behaviour 16

8.11 3.11 Logical Database Requirements 17

8.12 3.12 System Architecture 17

8.13 3.13 Layered Modular Design 18

9 Chapter 4: System Implementation 19

9.1 4.1 System Setup 19

9.2 4.1.1 Installation of Dependencies 19

9.3 4.2 Code Implementation 20

9.4 4.2.1 Webcam Feed Initialization 20

9.5 4.2.2 Face Detection and Landmark Identification 21

9.6 4.2.3 Drowsiness and Yawn Detection 21

3
[Link]. Contents Page No.

9.7 4.2.4 Lane Detection 22

9.8 4.2.5 Car Detection 22

9.9 4.2.6 Alerting Mechanism 23

9.10 4.3 Execution Flow 23

9.11 4.4 Handling Errors and Limitations 24

9.12 4.4.1 Lighting and Camera Issues 24

9.13 4.4.2 Real-Time Processing 24

9.14 4.5 Future Implementation Directions 25

9.15 4.6 Limitations 25

10 Chapter 5: Testing and Evaluation 26

10.1 5.1 Implementation Overview 26

10.2 5.1.1 Modules Executed 26

10.3 5.1.2 Dataflow Overview 27

10.4 5.2 Tools Used 27

10.5 5.3 Frontend Development 28

10.6 5.4 Drowsiness and Yawn Detection Implementation 28

10.7 5.5 Lane Detection Implementation 29

10.8 5.6 Alert Generation and Interface Feedback 30

10.9 5.7 Flowcharts and System Behavior 30

10.10 5.8 Actual Interface 31

11 Chapter 6: Conclusion and Future Work 32

11.1 6.1 Conclusion 32

4
[Link]. Contents Page No.

11.2 6.1.1 Key Achievements 32

11.3 6.2 Future Work 33

11.4 6.2.1 Integration of Deep Learning Models 33

11.5 6.2.2 Enhanced Vehicle and Object Detection 33

11.6 6.2.3 Improved Lighting and Environment 34

11.7 6.2.4 Optimization for Embedded Systems 34

11.8 6.2.5 Mobile and Cloud Integration 34

11.9 6.2.6 Advanced Sensor Integration 35

11.10 6.3 Final Thoughts 35

12 Chapter 7: References 36

5
 CHAPTER 1:

INTRODUCTION

1.1 INTRODUCTION

Road safety has become a pressing concern globally due to the rise in traffic and motorization.
Human error, especially fatigue and distraction, is among the top contributors to road accidents.
With more vehicles on the road, the need for proactive safety mechanisms has become crucial.
"Drive Alert" addresses this issue by implementing real-time monitoring and alerting
mechanisms for driver fatigue and distraction. It combines two major functionalities: Driver
Drowsiness and Yawn Detection, and Lane and Car Detection.
The core idea behind "Drive Alert" is to utilize computer vision techniques to ensure that the
driver remains alert and aware of the surroundings. Using tools like OpenCV, dlib, Haar
Cascades, and Mediapipe, this system processes live video feeds to detect signs of fatigue (such
as eye closure or yawning) and vehicular hazards (like lane departure or nearby cars). This dual
approach helps prevent accidents by providing timely alerts to the driver, making it a highly
effective tool for road safety.

This system has real-world applications in logistics, public transport, personal vehicles, and
rideshare services. Given its low computational requirements and ease of deployment, it can be
used in embedded systems or regular computing devices. The technology is scalable and
adaptable, making it suitable for both developed and developing regions facing the challenge of
road safety.

1.2 PURPOSE

The purpose of "Drive Alert" is to provide a robust, efficient, and real-time system that can
monitor a driver's level of alertness and vehicular context using computer vision. The project
addresses two primary concerns:

1. Driver Fatigue: Continuous monitoring of eye and mouth movement helps in

determining the drowsiness levels of a driver. This is done using EAR (Eye Aspect Ratio)
and MAR (Mouth Aspect Ratio) metrics derived from facial landmarks. The system can
detect prolonged eye closures and frequent yawning, triggering an alarm or alert system
when thresholds are breached.

2. Road Awareness: Through lane and vehicle detection using OpenCV's image processing
techniques and pre-trained Haar classifiers, the system identifies deviations in road
positioning or potential vehicle collisions.

6
This purpose aligns with global efforts to reduce road fatalities and make smart vehicles more
intelligent and user-protective. By ensuring that the driver stays alert and in-lane, the project
contributes to safer roads and fewer accidents.

1.3 PROJECT OBJECTIVES

• To detect driver drowsiness using eye-blink frequency and mouth opening.

• To calculate EAR and MAR values in real-time using dlib's facial landmark detection.

• To alert drivers with audio/visual signals when drowsiness or distraction is detected.

• To detect road lanes and provide feedback if a vehicle is deviating from its lane.
• To identify vehicles in close proximity using object detection with Haar cascade
classifiers.

• To integrate all functionalities into a single, user-friendly Python application.

• To optimize the system for performance using real-time webcam or video input.

1.4 PROJECT FLOW

The workflow of the Drive Alert system is modular and sequential. It includes:

1. Video Feed Capture: Capturing real-time video through webcam.

2. Face Detection: Using dlib’s face detector to locate the driver’s face.

3. Facial Landmark Extraction: 68-point landmarks used to identify eye and mouth
regions.

4. Drowsiness Logic: EAR and MAR values are computed to determine alertness.
5. Alert Trigger: Alarm activated if drowsiness is confirmed over multiple frames.

6. Lane Detection: Frames processed using Canny edges, ROI masking, and Hough Line
detection to identify lane boundaries.
7. Car Detection: Using pre-trained Haar cascade classifiers to identify and mark other
vehicles.
8. Result Display: GUI shows real-time status with overlays and bounding boxes.

1.5 PROJECT SCOPE

7
"Drive Alert" can be applied across various sectors and platforms:

• Commercial Fleet Vehicles: To monitor driver health and alertness.

• Public Transport: Buses and trains where operator fatigue is a safety issue.

• Personal Vehicles: Especially for individuals with long commutes or road trips.
• ADAS (Advanced Driver Assistance Systems): As a plug-and-play module in smarter
automotive systems.

• Embedded Systems: Raspberry Pi or Jetson Nano versions can be developed for portable
use.

1.6 PROBLEM DEFINITION

Road accidents due to driver fatigue account for thousands of deaths each year. Traditional alert
systems depend on physical feedback (like lane swerving), which is often too late to prevent an
accident. Similarly, cars without lane detection can swerve unknowingly into adjacent lanes,
leading to collisions. The need is for a proactive, vision-based monitoring system that alerts
drivers before a dangerous state escalates.

"Drive Alert" provides a solution by recognizing early indicators of drowsiness through eye and
mouth behavior, and preventing lane departure or vehicular collision through visual input
analysis.

1.7 MODULE OVERVIEW

1.7.1 Drowsiness and Yawn Detection Module

• Uses dlib's 68 facial landmarks to track eye and mouth.

• Computes EAR and MAR in each frame.

• Triggers alerts if thresholds (e.g., EAR < 0.25 for 30+ frames) are crossed.
1.7.2 Lane Detection Module

• Applies grayscale, Gaussian blur, Canny edge detection, and masking.

• Identifies probable lane lines using Hough Line Transform.

• Calculates center offset and provides visual feedback.

1.7.3 Car Detection Module

• Uses Haar cascade XML classifiers.
8
 • Detects nearby vehicles and outlines them with rectangles.

• Can be extended to object tracking and collision warnings.

1.8 IMPACT AND APPLICATION AREA

"Drive Alert" offers practical and wide-ranging benefits:

• Reduces risk of accidents due to driver drowsiness.

• Enhances awareness of road position and lane safety.

• Acts as a low-cost enhancement to vehicles without built-in ADAS.

• Can be integrated with smart dashboards and mobile platforms.

• Boosts confidence of fleet operators in ensuring driver safety.

CHAPTER 2:
THE OVERALL DESCRIPTION

2.1 PRODUCT PERSPECTIVE

"Drive Alert" is a standalone software product designed to act as an intelligent assistant for
vehicle safety by proactively monitoring the driver and the road environment using real-time
image processing. The product serves as a basic prototype for an Advanced Driver Assistance
System (ADAS), integrating visual analysis for both the internal (driver) and external (road)
environments. While traditional systems often rely on manual or post-event feedback
mechanisms, "Drive Alert" distinguishes itself by functioning in real-time and preemptively
alerting the driver to possible dangers.

This project does not require any proprietary hardware; a basic webcam and a computing device
with Python and OpenCV libraries suffice. It is built to be modular and extensible. Thus, it can
be integrated with existing vehicle dashboards or expanded with features like GPS tracking and
cloud logging in future iterations. In its current form, "Drive Alert" provides a solid foundation
for more advanced safety platforms.

2.1.1 USER INTERFACE

The user interface is kept minimal to avoid distracting the driver. Overlays are drawn directly on
the webcam feed, highlighting facial features, detected lanes, and nearby vehicles. Real-time
alerts for drowsiness are displayed via text warnings or audio beeps. The simplicity of the
interface ensures quick comprehension and efficient user interaction.

9
2.1.2 HARDWARE INTERFACES

The only required hardware interface is a webcam, which serves as the input device for capturing
the driver’s face and the front-road video feed. In future versions, support for infrared cameras or
night-vision modules can be added to improve accuracy during nighttime driving.

2.1.3 SOFTWARE INTERFACES

The backend of "Drive Alert" relies on Python-based libraries such as OpenCV, dlib, and Haar
cascades. System-level integration with operating systems is done via standard I/O and device
interfaces. The application is platform-independent and can run on any system supporting Python
and basic imaging libraries.
2.1.4 MEMORY CONSTRAINTS

The software is optimized to run in systems with 4GB RAM and minimal GPU usage. Real-time
frame processing is achieved without storing large image data, thereby keeping memory usage
low. All temporary computations are cleared frame-wise to avoid memory leaks.

2.2 PRODUCT FUNCTIONS

The product captures live video input, processes each frame to detect facial features, calculates
EAR and MAR to determine driver alertness, applies lane detection algorithms to monitor road
adherence, and detects surrounding vehicles. Alerts are triggered based on calculated thresholds.
The functionalities are modular, meaning each detection system works independently and
simultaneously.

2.3 USER CHARACTERISTICS

Users include vehicle operators, drivers, logistics supervisors, and tech developers. They are
expected to have basic knowledge of running software applications. The system is designed for
hands-free operation, requiring no manual input during driving. Setup and calibration are
minimal, making it suitable for non-technical users as well.
2.4 CONSTRAINTS

• Requires stable lighting conditions for accurate detection.

• Limited performance in poor weather or nighttime if basic webcams are used.

• System is dependent on real-time video processing, so hardware with lower

computational power may see frame lags.

• Not integrated with vehicle sensors (e.g., speedometer, GPS), which limits context-based
alerts.

10
 CHAPTER 3:

SPECIFIC REQUIREMENTS

3.1 REQUIREMENT SPECIFICATION

The "Drive Alert" system is a real-time driver monitoring and assistance system aimed at
reducing road accidents by proactively detecting driver drowsiness and road anomalies such as
lane deviation or nearby vehicle presence. This chapter outlines the specific requirements
necessary to design, build, and operate the system. These requirements include both hardware
and software specifications, as well as functional and non-functional goals to ensure effective
performance.
3.2 FUNCTIONAL REQUIREMENTS

Functional requirements define the specific behaviors and functionalities of the "Drive Alert"
system:
• Facial Feature Detection: Detect and extract facial landmarks in real-time using dlib’s
68-point facial landmark detector.
• Drowsiness Detection: Calculate Eye Aspect Ratio (EAR) and Mouth Aspect Ratio
(MAR) to determine fatigue levels.
• Alert Mechanism: Trigger an alarm (audio/visual) if the driver exhibits signs of
drowsiness or yawning.

• Lane Detection: Identify road lanes using Canny edge detection and Hough Line
Transform techniques.

• Vehicle Detection: Detect other vehicles in the scene using Haar Cascade classifiers.
• Frame Overlay: Display visual annotations such as lane boundaries, bounding boxes,
and alerts on the live feed.
• Multi-threaded Processing: Perform simultaneous detection tasks efficiently using
parallel processing.

3.3 NON-FUNCTIONAL REQUIREMENTS

These requirements focus on system quality and usability:

• Real-Time Processing: All computations should complete within the frame time (ideally
under 100ms per frame).
• Accuracy: High precision and recall for eye and lane detection to avoid false positives.
• User Friendliness: Minimal interface with easy startup and simple output presentation.

11
 • Portability: Should run on multiple platforms (Windows/Linux/Mac) with minimal
installation steps.

• Maintainability: Code should be modular and well-documented to support future

extensions.

• Reliability: System must operate continuously without crashes during prolonged use.

3.4 EXTERNAL REQUIREMENTS

External conditions and dependencies for optimal operation:

• Camera Input: A high-definition webcam with at least 30fps frame rate.

• Lighting Conditions: Sufficient ambient lighting or infrared sensors for accurate

detection.

• Processing Device: A computer with at least 2.0GHz processor and 4GB RAM.

• Python Runtime: Python 3.7+ and required libraries such as OpenCV, dlib, numpy.
3.5 USER REQUIREMENTS

The end-users include drivers, fleet operators, and technical teams. User expectations are:

• Simple Setup: Minimal configuration during first-time setup.

• Hands-Free Use: No manual intervention required while driving.

• Visual and Audio Alerts: Easy-to-understand alerts that do not distract from driving.

• Quick Feedback: Real-time alerts within 1-2 seconds of drowsiness or hazard detection.

3.6 HARDWARE REQUIREMENTS

• Processor: Intel i3/i5/i7 or equivalent AMD Ryzen (minimum 2.0 GHz).

• RAM: Minimum 4 GB.

• Webcam: At least 720p resolution with 30 FPS.

• Optional: Microphone and speaker for audio alerts.

3.7 SOFTWARE REQUIREMENTS

• Programming Language: Python 3.7 or higher.

• Libraries: OpenCV, dlib, imutils, numpy, scipy, threading.

• Development Environment: Visual Studio Code, Jupyter, or PyCharm.
• Operating System: Windows 10/11, Ubuntu 20+, or macOS.

12
3.8 SOFTWARE PRODUCT FEATURES

• Integrated Detection System: Combines multiple vision-based detection models in one

application.

• Real-Time Monitoring: Capable of real-time face, lane, and vehicle detection.

• Visual Feedback: Frame overlays indicating system observations.

• Cross-Platform Support: Runs on multiple operating systems with Python support.

• Modular Design: Enables addition/removal of components as needed.

3.9 MODULES

• Drowsiness Detection Module: Computes EAR/MAR to monitor eye closure and

yawning.

• Lane Detection Module: Detects and tracks lanes using edge detection and Hough
Transform.
• Vehicle Detection Module: Uses Haar cascade to identify vehicles in adjacent lanes.

• Alert Module: Provides real-time audio or visual cues to warn the driver.

3.10 SYSTEM BEHAVIOUR

• Captures continuous video feed from the webcam.

• Processes each frame to check for drowsiness signs and road layout.

• Detects anomalies such as eye closure or lane departure.

• Triggers alerts if thresholds are exceeded.

• Displays annotated output to inform users.

3.11 LOGICAL DATABASE REQUIREMENTS

Though "Drive Alert" operates in real-time, basic logging can be included:

• Session Logs: Store timestamps, duration, and event triggers (e.g., drowsiness detected).

• Alert Count: Log how many times an alert was triggered during a session.

• Performance Metrics: Store average EAR/MAR values and detection frame counts. Logs
can be stored in CSV or SQLite format for analysis.

3.12 SYSTEM ARCHITECTURE

The system architecture includes the following layers:

13
 • Input Layer: Webcam capturing live video feed.

• Processing Layer:

o Sub-modules for drowsiness, lane, and vehicle detection.

o EAR/MAR calculations, edge detection, classifier application.

• Alert Layer: Monitors thresholds and triggers alarms.

• Output Layer: Displays annotated frame with visual cues and text overlays.

3.13 LAYERED MODULAR DESIGN

The design follows a layered approach to allow modular testing and maintenance:

• Presentation Layer: Interface for the user (webcam feed + alert display).

• Application Layer: Logic to coordinate between modules and manage workflows.

• Detection Modules:

o Facial Analysis
o Lane Detection

o Object Detection

• Hardware Abstraction Layer: Interfaces with webcam and audio output. Each module
communicates via clearly defined interfaces, allowing individual modules to be upgraded
or replaced without affecting the entire system.

CHAPTER 4:
SYSTEM IMPLEMENTATION

4.1 SYSTEM SETUP

The "Drive Alert" system utilizes a variety of technologies to achieve real-time monitoring of
driver alertness and road safety. The foundation of the implementation lies in Python, a powerful
language for rapid development and integrating with libraries such as OpenCV, dlib, NumPy,
and Mediapipe. The following section explains the steps involved in setting up the system,
including the installation of essential dependencies.

4.1.1 Installation of Dependencies

Before running the system, certain libraries need to be installed. These libraries are essential for
the functionalities of "Drive Alert," from computer vision tasks like lane detection to facial

14
landmark analysis for detecting drowsiness. The required libraries and their installation steps are
as follows:

1. OpenCV: OpenCV is essential for image processing and computer vision tasks. It helps
in tasks like detecting faces, lanes, and other objects in the real-time video feed.

pip install opencv-python

2. dlib: dlib is used for facial landmark detection. By tracking 68 key facial points, we can
determine features such as the eye aspect ratio (EAR) and mouth aspect ratio (MAR) for
detecting drowsiness or yawning.

pip install dlib

3. NumPy: This library is used for numerical computations such as calculating ratios
between distances between facial landmarks.

pip install numpy

4. playsound: playsound is used to trigger sound alerts when drowsiness or any potential
hazards are detected.

pip install playsound

Additionally, the Haar Cascade classifier XML files for car detection need to be downloaded.
These files are pre-trained models used by OpenCV for object detection, including the
identification of vehicles in the camera feed.

After installing the necessary libraries, the Python environment is ready to run the system, which
is encapsulated in the [Link] file.

4.2 CODE IMPLEMENTATION

4.2.1 Webcam Feed Initialization

The system starts by capturing the real-time video feed from the webcam using OpenCV’s
[Link]() function. This captures each frame from the webcam, which will then be
processed in subsequent stages.

cap = [Link](0)

This line of code ensures that the webcam starts streaming video, and frames are continuously
read from the feed for further processing. The program will maintain an ongoing loop to process
every new frame.
4.2.2 Face Detection and Landmark Identification

15
The core of the system's alert mechanism lies in facial recognition and tracking. Using dlib, the
system detects faces in the captured frames and uses its facial landmark detection to identify key
points on the face. Specifically, it focuses on the eyes and mouth to calculate Eye Aspect Ratio
(EAR) and Mouth Aspect Ratio (MAR), which are essential for determining drowsiness and
yawning.

To detect faces, a Haar cascade classifier is used, followed by a dlib detector that identifies
faces within the frame:

face_cascade = [Link]([Link] +
'haarcascade_frontalface_default.xml')

detector = dlib.get_frontal_face_detector()

predictor = dlib.shape_predictor('shape_predictor_68_face_landmarks.dat')

The face detection process works by first detecting the face in the frame and then using the
predictor to identify the 68 facial landmarks. These landmarks represent key points on the face,
such as the eyes, nose, and mouth. By analyzing these, we can determine whether the driver is
exhibiting signs of drowsiness.

4.2.3 Drowsiness and Yawn Detection

Detecting drowsiness involves monitoring the Eye Aspect Ratio (EAR), which measures how
open or closed the eyes are. If the eyes are closed for too long (i.e., EAR falls below a certain
threshold), it is an indicator of drowsiness.

Similarly, the Mouth Aspect Ratio (MAR) measures how open the mouth is. A wide-open
mouth over an extended period can indicate yawning, another sign of fatigue.

def calculate_EAR(eye):

A = [Link](eye[1] - eye[5])

B = [Link](eye[2] - eye[4])

C = [Link](eye[0] - eye[3])

ear = (A + B) / (2.0 * C)
return ear

This function computes the EAR by calculating distances between key facial landmarks around
the eyes, and the MAR is calculated similarly for the mouth. If the EAR is below 0.25 for more
than 30 frames (a user-defined threshold), the system triggers an alert for drowsiness.

4.2.4 Lane Detection

16
Lane detection is essential for ensuring that the driver stays within the designated lane, reducing
the risk of lane departure accidents. The system applies several computer vision techniques to
detect lanes in the road:

1. Canny Edge Detection: This technique detects the edges in the image, which helps in
identifying the boundaries of the road and lanes.

2. Region of Interest (ROI) Masking: To avoid processing irrelevant parts of the image
(like the sky), the system applies an ROI mask that isolates the area where lanes are
likely to appear.

3. Hough Line Transform: This is used to detect straight lines in the image, particularly
the lane markings.

gray = [Link](frame, cv2.COLOR_BGR2GRAY)

edges = [Link](gray, 50, 150)

The edge-detected image is then processed to identify lane lines, providing visual feedback on
the vehicle’s position relative to the lane boundaries.
4.2.5 Car Detection

In addition to lane detection, the system also identifies nearby vehicles using Haar Cascade
Classifiers. These classifiers are pre-trained models that can detect objects like cars in images.
The classifier is applied to each frame, and bounding boxes are drawn around detected vehicles:

car_cascade = [Link]('[Link]')
The car detection system scans the frame for vehicles and marks them with rectangular boxes,
helping the driver be aware of nearby cars, especially in blind spots or when changing lanes.
4.2.6 Alerting Mechanism
Once the system detects that the driver is drowsy or if there’s a potential road hazard (such as
lane departure or an approaching vehicle), an alert is triggered. The alert can be both visual (text
warnings on the screen) and auditory (an alarm or sound played through the speakers).

if EAR < 0.25:

playsound("alert.mp3")

[Link](frame, "Drowsy!", (50, 50), cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 0, 255), 2)

The sound alert warns the driver, and a text message, such as "Drowsy!", appears on the screen,
clearly indicating the potential danger. The system ensures that the alerts are immediate and
visible, without overwhelming the driver.

17
4.3 EXECUTION FLOW

The "Drive Alert" system operates in real-time, constantly processing video frames to detect
various events. Here’s an overview of how the system executes:

1. Initialize Webcam: Start capturing video frames from the webcam.

2. Face Detection: Detect the driver’s face using dlib’s face detector.

3. Facial Landmark Extraction: Extract 68 key points on the face to calculate the EAR
and MAR.

4. Drowsiness Detection: If EAR or MAR drops below the threshold, an alert is triggered.

5. Lane Detection: Using edge detection and Hough lines, the system tracks the vehicle’s
lane.

6. Car Detection: Detect nearby vehicles and draw bounding boxes.

7. Alerting: If any of the conditions for alerting (drowsiness, lane deviation, car proximity)
are met, trigger a sound and visual warning.

8. Display the Result: Show the processed frame with overlays, such as bounding boxes
and text warnings.

4.4 HANDLING ERRORS AND LIMITATIONS

4.4.1 Lighting and Camera Issues

One limitation of the system is its performance under poor lighting conditions. Facial landmark
detection and lane identification can be less accurate if the driver is in dim lighting or if the
camera quality is low. To counter this, image preprocessing techniques like Gaussian Blur are
used to reduce noise in low-light conditions, and adjustments can be made for better camera
quality.

4.4.2 Real-Time Processing

The system processes video frames in real-time. On devices with limited processing power, such
as low-end laptops or embedded systems, frame drops may occur, leading to slight delays in
detection. Optimization techniques like reducing frame resolution or processing every other
frame can alleviate some of the load.

4.5 FUTURE IMPLEMENTATION DIRECTIONS

The system can be expanded and enhanced in several ways:

18
 1. Deep Learning: Integrating deep learning models like Convolutional Neural Networks
(CNNs) could improve detection accuracy for vehicles and lanes.

2. Night Driving Support: Incorporating infrared cameras or using night-vision technology

can improve detection in low-light conditions.

3. Integration with Vehicle Systems: Integrating with onboard sensors such as

speedometers, GPS, and accelerometers would allow the system to offer more contextual
alerts based on the vehicle's speed, position, and road type.
4. Mobile and Embedded Versions: Developing mobile applications or versions optimized
for embedded devices like Raspberry Pi could make "Drive Alert" more versatile.

4.6 LIMITATIONS

• Weather and Lighting Conditions: The system’s performance might degrade in adverse
weather conditions like rain or fog or under poor lighting.

• Hardware Limitations: Low-end devices might struggle to process the video feed in real
time, leading to frame delays.

• Vehicle System Integration: The system does not currently integrate with other vehicle
sensors, limiting its ability to offer context-specific warnings.

CHAPTER 5:

TESTING AND EVALUATION

5.1 IMPLEMENTATION OVERVIEW

The implementation of the "Drive Alert" project is focused on creating a real-time driver
monitoring system that integrates computer vision techniques to detect drowsiness and yawning
symptoms as well as identify unsafe lane drifting behavior. The system is developed using
Python and OpenCV, incorporating essential components such as facial landmark detection, head
pose estimation, and image processing algorithms for lane detection. All these functionalities are
implemented efficiently in a single script ([Link]) for seamless execution. The application
provides visual alerts and audio warnings, ensuring immediate driver feedback during potentially
dangerous conditions.

5.1.1 MODULES EXECUTED

19
The core implementation of the Drive Alert system revolves around the execution of two main
modules:

• Drowsiness and Yawn Detection Module: Uses Dlib and OpenCV to detect facial
landmarks and compute Eye Aspect Ratio (EAR) and Mouth Aspect Ratio (MAR).
Triggers an alarm if prolonged eye closure, yawning, head tilting, or absence of face is
detected.

• Lane Detection Module: Utilizes Canny edge detection, region of interest masking, and
Hough Transform to detect lane lines and provide alerts when the vehicle veers off the
lane.

Each module functions independently but runs concurrently to offer comprehensive driver
monitoring.

5.1.2 DATAFLOW OVERVIEW

1. Camera Input: Video is captured through the webcam in real time.

2. Face Detection: Dlib detects the driver’s face and extracts facial landmarks.

3. EAR and MAR Calculation: Eye and mouth ratios are calculated frame-by-frame to
detect drowsiness or yawning.

4. Head Pose Detection: Checks head tilt using facial angles.

5. Lane Detection Pipeline: Converts frame to grayscale, applies Gaussian blur, and uses
Canny edge detection, ROI masking, and Hough Transform.

6. Alerts Generation: Based on EAR/MAR/head pose and lane deviation, audio alerts and
on-screen messages are triggered.
7. Visualization: OpenCV displays processed frames with overlays indicating status.

5.2 TOOLS USED

• Python 3.8+ – Core programming language

• OpenCV – For image and video processing

• Dlib – Facial landmark detection

• Imutils – Utility functions for image transformation

• Playsound / Pygame – For playing alert sounds

20
 • Numpy – Numerical operations on arrays

• Matplotlib (for debugging) – To visualize processing results

• Laptop Webcam – Hardware for capturing real-time video

These tools ensure a lightweight and efficient implementation capable of running on regular
hardware.

5.3 FRONTEND DEVELOPMENT

The frontend of the "Drive Alert" system primarily revolves around real-time visual feedback,
which is rendered directly from the webcam feed and processed through OpenCV. The visual
overlay includes bounding boxes, landmark indicators, eye aspect ratio scores, and gesture hints
to guide the user. While the system doesn't rely on a traditional GUI-based frontend (like
HTML/CSS interfaces), its visual interface is an essential part of user interaction.

Key frontend features include:

• Live Webcam Feed: Displayed using OpenCV’s imshow function.

• Overlay Indicators: Shapes and text such as circles for eye landmarks, polygons for the
mouth, and gesture status messages.

• Alert Popups: On-screen alerts when drowsiness or yawning is detected.

21
5.4 DROWSINESS AND YAWN DETECTION IMPLEMENTATION

This module is the core of the Drive Alert system. It uses facial landmark detection to monitor
the driver’s eye movements, yawning behavior, and head posture. The system calculates the Eye
Aspect Ratio (EAR) and Mouth Aspect Ratio (MAR) in real-time to detect signs of fatigue.

Step-by-Step Breakdown:
1. Facial Detection:

o Uses Dlib’s 68-point facial landmark detector.

o Ensures face is located before continuing detection.

22
2. Eye Aspect Ratio (EAR):

o EAR is computed using the distance between eye landmarks.

o A threshold (typically 0.25) is used to detect closed eyes.

o If the eyes remain closed for more than a defined frame count (e.g., 40), an alarm
is triggered.

3. Mouth Aspect Ratio (MAR):

o Similar to EAR, calculated using vertical and horizontal distances of mouth

landmarks.

o Detects yawning based on a MAR threshold (e.g., > 0.7).

23
 4. Head Tilt Detection:

o Head tilt is inferred using nose bridge landmarks.

o If head angle exceeds certain bounds, it may indicate sleepiness.

5. Face Absence Detection:

o If no face is detected for a few seconds, an alert is triggered assuming

inattentiveness.

6. Alarm System:

o Audio alerts are played using Playsound or Pygame.

o Message overlays appear on-screen: “Drowsiness Detected”, “Yawning Alert”,

etc.

7. Visualization:

o EAR/MAR values, eye contours, and alert messages are shown on video frames.

This robust multi-parameter approach improves the accuracy of detecting drowsy or distracted
driving.

5.5 LANE DETECTION IMPLEMENTATION

24
The lane detection module processes the input video frame from the webcam to detect the lane
markings on the road in real-time. It helps alert the driver when the vehicle starts drifting away
from the detected lane, thus preventing accidents due to unintentional lane departures.

Key Steps:

1. Frame Preprocessing:

o Convert each frame to grayscale.

o Apply Gaussian blur to reduce noise.

2. Edge Detection:

o Use Canny edge detector to find edges in the image.

3. Region of Interest (ROI):

o A polygonal mask is applied to focus only on the part of the image where lane
lines are expected.
4. Line Detection:

o Apply the Hough Line Transform to detect straight lines.

5. Lane Line Drawing:

o Identified lines are drawn over the original frame using OpenCV.

6. Deviation Alert:

o If the vehicle is detected veering outside of the expected lane center, an alert is
issued.

5.5.1 IMPLEMENTATION DIAGRAM

25
5.6 ALERT GENERATION AND INTERFACE FEEDBACK

• Visual Alerts:

o Text overlays such as “LANE DRIFT ALERT” or “DROWSINESS

DETECTED”.

• Audio Alerts:

o Immediate sound cues triggered using playsound module.

• Frame Updates:
o Frames are updated in real-time with visual indicators.

This layered and modular approach ensures that each condition—whether it is drowsiness,
yawning, or lane deviation—is detected and handled independently and reliably.

5.7 FLOWCHARTS AND SYSTEM BEHAVIOR

Lane Detection Flowchart:

26
Drowsiness/Yawning Detection Flowchart:

27
→ Evaluate Thresholds (EAR < 0.25, MAR > 0.7)

→ If Threshold Breached for N Frames → Trigger Alert

→ Display Alerts & Landmarks

→ Loop to Next Frame

These flowcharts simplify and visualize how data moves and decisions are made at each step.

5.8 Actual Interface

28
29
30
 CHAPTER 6:
CONCLUSION AND FUTURE WORK
31
6.1 CONCLUSION

"Drive Alert" is a comprehensive real-time safety system designed to monitor driver fatigue, lane
adherence, and vehicle proximity using computer vision techniques. The project successfully
integrates multiple advanced technologies such as drowsiness detection, lane detection, and
vehicle detection to create a system that can enhance road safety by alerting drivers to potential
hazards. The core idea is to provide proactive and immediate feedback to drivers, helping them
avoid accidents caused by fatigue, distraction, or lane departure.

The system performs well under most driving conditions, accurately detecting fatigue indicators
like eye closures and yawning, as well as identifying lane boundaries and surrounding vehicles.
The use of facial landmark detection for driver monitoring, coupled with traditional image
processing techniques like Canny edge detection and Hough Line Transform for lane detection,
proves to be an effective approach for real-time road safety applications.
Testing results showed that the system performed well in normal driving conditions, but
challenges such as poor lighting, extreme weather, and low camera quality impacted the system's
performance. Despite these limitations, the system is capable of delivering real-time alerts for the
most common road hazards, and it demonstrates significant potential for real-world applications
in improving road safety.

The integration of computer vision with embedded systems, using widely available hardware like
webcams and embedded devices such as the Raspberry Pi, further enhances the accessibility
and practicality of the system. Its modular nature allows for easy customization and future
expansion, making it adaptable for use in a wide range of vehicles, from personal cars to
commercial fleets.

"Drive Alert" addresses key issues in road safety, providing a reliable, cost-effective solution that
can be deployed across different sectors, including public transportation, logistics, and private
vehicles. The system's use of open-source tools and its ability to run on low-resource devices
also open the door to widespread adoption, especially in developing regions where road safety
concerns are high, and advanced driver assistance systems (ADAS) are often unavailable.

6.1.1 Key Achievements

The major achievements of the "Drive Alert" system can be summarized as follows:

• Accurate Drowsiness Detection: The system effectively monitors driver fatigue through
the use of Eye Aspect Ratio (EAR) and Mouth Aspect Ratio (MAR) metrics, providing
early warnings before the driver reaches dangerous levels of drowsiness.

• Real-Time Lane Detection: The system uses Canny edge detection and Hough Line
Transform to detect lane boundaries, giving drivers a visual cue if they deviate from their
lane, which helps in reducing lane departure accidents.

32
 • Vehicle Detection and Collision Avoidance: By employing Haar Cascade Classifiers,
the system detects nearby vehicles, offering warnings about potential collisions and
helping drivers maintain a safe distance.

• Low-Cost and Accessible: The system operates with minimal hardware requirements,
making it suitable for deployment in various vehicle types and affordable for individual
users and businesses alike.

• Modular and Scalable: The modular design of the system makes it adaptable for future
enhancements, such as GPS tracking, cloud integration, or the addition of advanced
sensors for improved performance.

6.2 FUTURE WORK

While the "Drive Alert" system has proven to be effective in many aspects of driver safety, there
are several opportunities for further enhancement and optimization that could be explored in
future versions of the system. These improvements could increase the system’s robustness,
expand its functionality, and ensure better performance under challenging conditions. Below are
some potential areas for future work:

6.2.1 Integration of Deep Learning Models

One of the key areas for future improvement is the integration of deep learning techniques to
enhance the accuracy and robustness of the system, particularly in object detection and lane
detection. While the current system uses traditional computer vision methods like Haar
Cascade Classifiers and Canny edge detection, deep learning models, such as Convolutional
Neural Networks (CNNs), could significantly improve detection accuracy, especially in
complex and cluttered environments.

Deep learning models have been proven to outperform traditional methods in tasks like object
detection and classification. By integrating a pre-trained CNN model for vehicle detection, for
example, the system could detect a wider range of vehicles, including motorcycles, bicycles, and
pedestrians, thus enhancing safety further. Additionally, deep learning-based lane detection
algorithms, such as LaneNet or ENet, could improve the accuracy of lane boundary detection,
especially on roads with faded or missing lane markings.

6.2.2 Enhanced Vehicle and Object Detection

In its current form, "Drive Alert" uses Haar cascade classifiers for detecting vehicles. However,
this method can sometimes struggle in highly dynamic environments, such as busy urban streets
or highways, where vehicles move quickly or are obstructed by other objects. Future work could
focus on improving vehicle detection accuracy by utilizing more advanced models like YOLO
(You Only Look Once) or Faster R-CNN, which have proven to be highly efficient and
accurate in real-time object detection tasks.

33
Furthermore, the system could be expanded to detect other objects on the road, such as
pedestrians, cyclists, road signs, and traffic lights. This would provide a more comprehensive
safety system capable of alerting the driver to various potential hazards, including pedestrians
crossing the road or approaching intersections with traffic lights.
6.2.3 Improved Lighting and Environmental Adaptability

One of the major challenges identified during testing was the system's reduced performance in
low-light conditions. In particular, facial landmark detection becomes less accurate when the
driver’s face is poorly lit. This limitation could be addressed by integrating infrared (IR)
cameras or night vision modules, which are commonly used in advanced driver assistance
systems (ADAS) to detect the driver’s face even in the dark.

Similarly, the lane detection system could be improved to work effectively in poor weather
conditions such as fog, rain, or snow, where visibility is compromised, and road markings may
be obscured. Adding weather-adaptive algorithms, like those used in Rain or Fog Detection
Systems, could help the system adjust to changing conditions and maintain performance.
6.2.4 Performance Optimization for Embedded Systems

Although the system works well on standard laptops, its performance on low-resource devices
like the Raspberry Pi could be further optimized. During testing, the Raspberry Pi 4 struggled to
maintain high frame rates when processing real-time video feeds, especially when running the
full set of features simultaneously. Future versions of the system could optimize its performance
by:

• Reducing the resolution of video feeds and processing fewer frames per second.

• Offloading intensive tasks (such as object detection) to cloud-based servers for faster
processing.

• Using hardware acceleration (e.g., GPU or TPU support) to speed up image processing
tasks.

6.2.5 Mobile and Cloud Integration

To make the system more accessible and portable, future work could involve integrating "Drive
Alert" into mobile platforms. This would enable the system to run on smartphones, which are
increasingly equipped with high-quality cameras and powerful processors. By leveraging the
cloud, the system could offload intensive tasks like object detection and lane recognition to
remote servers, thereby improving performance on mobile devices.
Mobile integration could also enable additional features such as GPS tracking and cloud-based
logging, allowing users to track their driving habits, receive feedback over time, and potentially
share their data with fleet managers or insurance companies.

34
6.2.6 Advanced Sensor Integration

Finally, the integration of additional sensors such as LiDAR, radar, and ultrasonic sensors
could further improve the accuracy of the system, especially for vehicle detection and collision
avoidance. These sensors can provide more detailed information about the surrounding
environment, such as the precise distance to nearby objects or vehicles, which would improve the
system’s ability to detect hazards and make real-time decisions.

6.3 FINAL THOUGHTS

In conclusion, the "Drive Alert" system is a promising step forward in improving road safety by
addressing key issues such as driver fatigue, lane departure, and collision risks. With further
refinement, including the integration of deep learning, enhanced environmental adaptability, and
mobile support, the system has the potential to significantly impact road safety across a wide
range of vehicle types and driving conditions. As technology continues to evolve, so too can the
capabilities of systems like "Drive Alert," making roads safer and more intelligent for everyone.

CHAPTER 7:

REFERENCES

In any research or project, references are crucial as they provide a foundation for the concepts,
methods, and technologies used in the work. Below is a list of key resources and references that
contributed to the development of the "Drive Alert" system. These references span academic
papers, tutorials, documentation, and online courses that provided the necessary theoretical
knowledge and practical techniques for implementing the various components of the system.

7.1 ACADEMIC AND TECHNICAL PAPERS

1. Jiang, H., & Li, Z. (2018). Driver Drowsiness Detection Based on Facial Features.
Journal of Computational and Theoretical Nanoscience, 15(3), 1037–1042.

o This paper discusses methods for detecting driver drowsiness using facial features
and eye movement. The EAR (Eye Aspect Ratio) used in the "Drive Alert" system
is based on the concepts proposed in this paper.

2. Smith, S., & Anderson, M. (2017). A Survey of Vision-Based Driver Assistance Systems.
International Journal of Computer Vision, 22(1), 58–75.

o This paper provides an in-depth survey of various computer vision-based driver

assistance systems, including lane detection and object detection, which form the
basis of the "Drive Alert" system.

35
 3. Xu, X., & Zhang, Y. (2019). Real-Time Lane Detection and Tracking for Autonomous
Vehicles. IEEE Transactions on Intelligent Transportation Systems, 25(5), 1321–1329.

o This work focuses on lane detection algorithms, including edge detection

techniques like Canny and Hough Transform, which are implemented in the
"Drive Alert" system.

4. Zhao, H., & Liu, X. (2020). Vehicle Detection Using Haar Cascade Classifiers. Journal
of Computer Vision and Image Processing, 34(6), 215–223.
o The methodology for detecting vehicles using Haar cascade classifiers was a
critical reference for the vehicle detection module in "Drive Alert."
7.2 ONLINE RESOURCES AND TUTORIALS

1. PyImageSearch (2023). Real-Time Drowsiness Detection with Python and OpenCV.

Retrieved from [Link]
o PyImageSearch is a well-known platform for tutorials on computer vision. The
real-time drowsiness detection tutorial provided insights into implementing EAR
and MAR metrics for detecting driver fatigue.

2. OpenCV Documentation. Canny Edge Detection. Retrieved from

[Link]

o The OpenCV documentation provided the foundation for implementing Canny

edge detection, a key part of the lane detection algorithm used in "Drive Alert."

3. dlib Documentation. Face Detection and Landmark Detection with dlib. Retrieved from
[Link]

o dlib’s documentation was used to understand how to implement facial landmark

detection for monitoring driver alertness, which is an essential component of the
system.

4. Medium (2021). How to Detect Cars with OpenCV’s Haar Cascade Classifiers.
Retrieved from [Link]

o This tutorial provided practical steps for using Haar cascade classifiers to detect
vehicles, which was implemented in the "Drive Alert" system to prevent
collisions.
7.3 SOFTWARE LIBRARIES AND TOOLS

1. OpenCV (2023). Open Source Computer Vision Library. Retrieved from

[Link]

36
 o OpenCV is the main library used in the "Drive Alert" system for image processing
tasks like edge detection, lane detection, and face detection.

2. dlib (2023). dlib C++ Library. Retrieved from [Link]

o dlib is used in the project for its robust facial landmark detection capabilities,
which are critical for monitoring the driver’s alertness and detecting drowsiness.

3. Haar Cascade Classifier (2023). Vehicle Detection with Haar Cascades. Retrieved from
[Link]

o The Haar cascade classifiers from OpenCV are used to detect vehicles in the
system. The pre-trained XML files provided by OpenCV are leveraged for real-
time vehicle detection.

4. Mediapipe (2023). Mediapipe: Cross-Platform Framework for Building Multimodal

Applied ML Pipelines. Retrieved from [Link]
o Mediapipe is used in the "Drive Alert" system for tasks such as detecting facial
landmarks and tracking head movements, which play a role in drowsiness
detection and road awareness.

7.4 BOOKS AND GUIDES

1. Raj, R. (2021). Python for Computer Vision: Master OpenCV with Python in 4 Hours.
Packt Publishing.

o This book provided detailed guidance on how to use OpenCV for real-time
computer vision tasks such as object detection and lane tracking, which were
implemented in the "Drive Alert" system.

2. Gonzalez, R. C., & Woods, R. E. (2018). Digital Image Processing (4th Edition).
Pearson Education.

o This book is a seminal resource for understanding image processing techniques

like edge detection and thresholding, which are fundamental to the lane and
vehicle detection modules in "Drive Alert."

3. Kothari, P., & Khanduja, V. (2019). Machine Learning and Computer Vision in
Intelligent Transport Systems. Wiley-Blackwell.

o This book provided valuable insights into the integration of machine learning and
computer vision for transportation applications, which helped in understanding
the broader context of the "Drive Alert" system.
4. Sharma, A. (2020). Practical Python and OpenCV: Intelligent Computer Vision Projects
with Python. Apress.

37
 o A practical guide for implementing real-time computer vision projects with
Python and OpenCV, this book served as an excellent resource for understanding
the application of Python in building the "Drive Alert" system.

7.5 OTHER RESOURCES

1. GitHub Repository for "Drive Alert" Project:

o The GitHub repository houses all the code, configuration files, and dependencies
used in the "Drive Alert" system, providing a public and reproducible resource for
others to implement or expand on the project.

o GitHub Repository

2. YouTube Tutorial by Sentdex (2020). Face and Eye Detection Using Python and
OpenCV. Retrieved from [Link]

o Sentdex’s tutorial on face and eye detection helped in understanding the

application of Haar Cascade Classifiers and dlib facial landmarks to track
facial movements, particularly for detecting drowsiness in drivers.
3. Stack Overflow Community: Various Discussions on Computer Vision, OpenCV, dlib,
and Python Integration. Retrieved from [Link]
o The Stack Overflow community was an invaluable resource for troubleshooting
issues and finding solutions for problems encountered during the development of
"Drive Alert," including integration issues, performance optimizations, and
algorithm tuning.

4. T. Soukupová and J. Čech, “Real-Time Eye Blink Detection Using Facial Landmarks,”
Center for Machine Perception, Czech Technical University, 2016.

5. P. Viola and M. Jones, “Robust Real-Time Face Detection,” International Journal of

Computer Vision, 2001.

6. Redmon, J. and Farhadi, A. “YOLOv3: An Incremental Improvement,”

arXiv:1804.02767, 2018.

7. OpenCV Documentation – [Link]

8. Dlib Library Documentation – [Link]

9. Python Software Foundation – [Link]

10. “Deep Learning for Vehicle Detection and Tracking,” Towards Data Science,
[Link]

38
11. cvlib: a simple, high-level, easy-to-use open-source library for object detection –
[Link]

12. Raspberry Pi Foundation – [Link]

13. MediaPipe by Google – [Link]

14. Documentation for Darknet/YOLO – [Link]

15. Scikit-image and Scikit-learn Libraries – [Link] and [Link]

[Link]

39