Made byBobr AI

Driver Drowsiness Detection using Computer Vision & ML

Learn how to build a real-time driver drowsiness detection system using Mediapipe, Eye Aspect Ratio (EAR), and Python for improved road safety.

#computer-vision#machine-learning#python#mediapipe#opencv#road-safety#ai-project
Watch
Pitch
College Logo

A Mini Project Report

DRIVER DROWSINESS DETECTION

USING COMPUTER VISION

Department of Computer Science & Engineering

Barak Valley Engineering College

(Affiliated to ASTU, Approved by AICTE)

Nirala, Sribhumi, Assam - 788701

May 2026

STATUS: ACTIVE

Submitted by: Baharul Islam | Bhaskar Sha | Bongso Hanse | Hafizur Rahman | Md. Dilshan Ul Hassan Choudhury

Made byBobr AI

TABLE OF CONTENTS

02

College Logo
01. Introduction
02. Need of Machine Learning
03. Literature Survey
04. Existing Technology
05. Problem Statement
06. Objectives
07. Requirement Analysis
08. Data Flow Diagram
09. Use Case Diagram
10. Methodology
11. Eye Aspect Ratio (EAR)
12. Classification Model
13. Dataset Description
14. Experimental Setup
15. Results & Discussion
16. Performance Evaluation
17. Limitations
18. Conclusion
19. Future Scope
20. References
Made byBobr AI
College Logo
03

INTRODUCTION

Road accidents are a major global concern — driver drowsiness is a leading cause.
Fatigue, lack of sleep, and long driving hours reduce alertness and reaction time.
Traditional vehicle safety systems monitor external conditions only (braking, collision avoidance) — NOT the driver's internal state.
Need for intelligent real-time systems to detect early signs of drowsiness.
This project proposes a Driver Drowsiness Detection System using Computer Vision and Machine Learning.
Core technique: Eye Aspect Ratio (EAR) computed via Mediapipe Face Mesh facial landmarks.
Lightweight, real-time, cost-effective — no GPU required.

~20% of fatal accidents caused by drowsy driving

Made byBobr AI
04
Logo

NEED OF MACHINE LEARNING

Feature Extraction

Feature Extraction

EAR converts visual data into numerical form for analysis

Binary Classification

Binary Classification

Classifies driver as Alert (0) or Drowsy (1)

Decision Boundary

Decision Boundary

EAR < 0.21 threshold acts as ML decision boundary

Temporal Analysis

Temporal Analysis

Frame count distinguishes blinking from drowsiness

Scalability

Scalability

Extendable to SVM, Logistic Regression, CNN models

💡 ML enables adaptability across different users, lighting, and driving conditions.

Made byBobr AI
05

LITERATURE SURVEY

Logo
1
Paper 1

Real-Time Eye Blink Detection using Facial Landmarks

Introduces EAR concept. EAR remains constant when eyes open, drops when closed. Efficient for real-time use.

2
Paper 2

Driver Drowsiness Detection Using Computer Vision

Uses OpenCV webcam-based eye tracking. Non-intrusive, cost-effective.

3
Paper 3

Drowsy Driver Detection Using ML

SVM and neural networks for classification. High accuracy but needs large datasets.

4
Paper 4

Real-Time Fatigue Detection Using Mediapipe

Mediapipe Face Mesh for accurate landmarks. Faster than Haar cascades.

5
Paper 5

Vision-Based Driver Monitoring System

Combines eye detection + head movement. Multi-feature reduces false detection.

Made byBobr AI
Logo
06

EXISTING TECHNOLOGY

OpenCV Icon

OpenCV

Open Source Computer Vision Library

  • Video capture from webcam
  • BGR to RGB frame conversion
  • Image flipping, resizing, drawing
  • Real-time visualization with overlays
Lightweight • Cross-platform • Python-ready
Mediapipe Icon

Mediapipe Face Mesh

Google's Real-Time Perception Framework

  • 468 facial landmark detection
  • High-speed, low-latency processing
  • Accurate eye tracking
  • CPU-only, no GPU needed
Accurate • Fast • Cross-platform
ML Icon

ML Techniques

Classification & Feature Analysis

  • Logistic Regression — binary classification
  • Support Vector Machine (SVM)
  • Neural Networks / Deep Learning
  • Feature-based EAR classification
Scalable • Adaptable • Efficient
Made byBobr AI
07

PROBLEM STATEMENT

Logo
Warning
"

Most vehicle safety systems monitor roads, NOT the driver.

"
1

Lack of Real-Time Monitoring

Traditional systems don't continuously track driver alertness.

2

Inability to Detect Early Signs

Subtle behavioral changes like slow blinking go undetected.

3

High Cost of Advanced Systems

Infrared/EEG-based solutions are expensive, limited to premium vehicles.

4

Dependency on Manual Observation

Drivers may not recognize their own fatigue.

5

Limited Adaptability & Accuracy

Fixed rules fail across users, lighting, and environments.

6

No Lightweight Real-Time Solutions

Deep learning models need powerful hardware, impractical for real-time use.

This project addresses all the above challenges with a lightweight, real-time, cost-effective CV-based solution.

Made byBobr AI
08
Logo

OBJECTIVES OF THE PROJECT

01 — Real-Time Detection

Develop a real-time drowsiness detection system using webcam input.

02 — ML Classification

Apply Machine Learning concepts for binary classification (Alert / Drowsy).

03 — Eye Closure via EAR

Detect eye closure using Eye Aspect Ratio (EAR) feature extraction.

04 — Instant Audio Alert

Trigger immediate sound alarm (Pygame) when drowsiness is detected.

05 — Low-Cost Solution

Create a lightweight system deployable on standard CPU hardware.

Under the Guidance of: Mr. Swmdwm Basumatary, Asst. Professor, Dept. of CSE, BVEC

Made byBobr AI
09

Requirement Analysis

Logo

Functional Requirements

Check
Capture Video Input
Continuous live webcam feed, stable frame acquisition
Check
Detect Face & Eyes
Mediapipe Face Mesh for facial landmark localization
Check
Calculate EAR
Compute Eye Aspect Ratio per frame using 6 eye landmark points
Check
Classify Driver State
Alert / Drowsy / Sleeping classification based on EAR + frame count
Check
Trigger Alert
Audible alarm + visual warning when drowsiness detected

Non-Functional Requirements

Speed
Real-Time Performance
20–30 FPS processing, no noticeable delay
Target
High Accuracy
Minimize false positives (normal blinks) and false negatives (missed drowsiness)
Stopwatch
Low Latency
Fraction-of-a-second response from frame capture to alert trigger
Note: System must handle variations in head position, lighting, and partial occlusions.
Made byBobr AI
10

DATA FLOW DIAGRAM (DFD)

Logo

Level 0 — Context Diagram

Driver
Facial video (webcam)
Drowsiness Detection
System
Drowsiness alert
Alert Output

Level 1 — Detailed Flow

Driver — Live video stream
P1: Webcam Input & Frame Capture — Continuous frame extraction
P2: Face Detection (Mediapipe) — Facial landmark localization
P3: Eye Landmark Extraction — 6 eye region points per eye
P4: EAR Calculation — Eye openness ratio
P5: ML Classification — Drowsy vs Alert decision
Alert Trigger — Alarm/buzzer
Normal State — Continue monitoring
Made byBobr AI
11

USE CASE DIAGRAM

College Logo
Actor: Driver
Passive external entity monitored via webcam
Capture VideoWebcam captures real-time driver frames
Detect FaceMediapipe identifies facial landmarks
Extract Eye LandmarksEye region coordinates selected
Calculate EAREye Aspect Ratio computed per frame
Detect DrowsinessML logic classifies Alert or Drowsy
Trigger AlertAlarm activated on drowsiness detection
Note:All use cases are linked with «include» dependency relationships
«include» «include» «include» «include» «include»
Drowsiness Detection System
Driver Actor
Driver
Capture Video
Detect Face
Extract Eye Landmarks
Calculate EAR
Detect Drowsiness
Trigger Alert
Made byBobr AI
Logo
12

DETAILED METHODOLOGY

1

Data Acquisition

  • Webcam captures live video
  • Frame-by-frame real-time extraction
  • Non-intrusive, continuous
2

Feature Extraction

  • Mediapipe detects 468 facial landmarks
  • Eye region coordinates extracted
  • EAR computed (6 points per eye)
3

Feature Processing

  • EAR compared to threshold (0.21)
  • Frame counter incremented/reset
  • Noise filtered (blink vs drowsiness)
4

Classification

  • Rule-based ML binary classifier
  • EAR < 0.21 AND frames > 30 → Drowsy
  • Otherwise → Alert
5

Output & Alert

  • Audio alarm via Pygame
  • Visual warning on screen
  • System continues monitoring
ear_computation.py
EAR = ( ||p2-p6|| + ||p3-p5|| ) / ( 2 × ||p1-p4|| )
Made byBobr AI
13

EYE ASPECT RATIO (EAR)

What is EAR?

EAR measures eye openness using 6 facial landmark points per eye.

EAR = (||p2-p6|| + ||p3-p5||) / (2 × ||p1-p4||)

p1, p4: horizontal width

p2, p3, p5, p6: vertical pairs

Eyes OPEN → EAR ≈ 0.30–0.45

Eyes CLOSED → EAR < 0.21

Eye Diagram
p1
p2
p3
p4
p5
p6

EAR Threshold Visualization

0.00
Drowsy
Transition
0.45
Alert
Threshold = 0.21
0.21
0.30
State EAR Range
Alert 0.30 – 0.45
Transition 0.21 – 0.30
Drowsy < 0.21

Continuous frames with EAR < 0.21 exceeding acceptable blink duration trigger a systemic drowsiness alert.

Made byBobr AI
14

CLASSIFICATION MODEL

Rule-Based ML Classifier

Binary classification using EAR + frame count as features

IF EAR < 0.21 AND frame_count > 30:
→ Class 1: DROWSY
ELSE:
→ Class 0: ALERT

Binary Classes

0 = ALERT — Eyes open, normal
1 = DROWSY — Prolonged eye closure

Temporal Analysis

Normal blink: EAR drops briefly → counter resets
Drowsiness: EAR stays low > 30 frames → alert triggered
Frame counter prevents false positives
Logo
YES NO YES NO
EAR Value Computed
EAR < 0.21?
Increment Frame Counter
Frame Count > 30?
Reset Frame Counter
TRIGGER ALERT Drowsy
Continue Monitoring
Made byBobr AI

15

DATASET & EXPERIMENTAL SETUP

College Logo

Dataset Description

No predefined dataset used — real-time dynamic data from webcam.

Primary Feature: EAR (Eye Aspect Ratio) computed per frame

Frame No. EAR Value Frame Count State
1 0.32 0 Alert
2 0.29 0 Alert
3 0.20 5 Transition
4 0.18 30 Drowsy

Data generated dynamically — no storage needed → lightweight system

Experimental Setup

  • 🐍 Python 3.x — Core Programming Language
  • 👁 OpenCV — Video capture & image processing
  • 🧠 Mediapipe — Face mesh landmark detection
  • 🔢 NumPy — Numerical computations
  • 🔊 Pygame — Audio alert system
  • 💻 Standard CPU — No GPU required
  • 📷 Webcam — Real-time video input
🚀

Performance: 20–30 FPS processing speed

Made byBobr AI
College Logo

16

RESULTS & DISCUSSION

Case 1: Alert State

Open Eye

Eyes Fully Open

EAR Value: 0.32
Frame Count: 0
✅ GREEN ALERT — No action taken

System continues monitoring

Case 2: Normal Blink

Half Closed Eye

Brief Eye Closure

EAR Value: 0.18
Frame Count: 3 (low)
🟡 AMBER TRANSITION — No false alert

Counter resets — blink detected correctly

Case 3: Drowsy State

Closed Eye

Prolonged Eye Closure

EAR Value: 0.16
Frame Count: 32+
🔴 RED DROWSY — Alert Triggered!

Audio alarm activated immediately

System accurately distinguishes blinking from drowsiness with real-time alert response.

Made byBobr AI

17

PERFORMANCE EVALUATION

College Logo

Confusion Matrix

Predicted Alert Predicted Drowsy
Actual Alert 45 (TP) 5 (FP)
Actual Drowsy 4 (FN) 46 (TN)

Accuracy = (TP + TN) / Total = (45+46)/100 ≈ 91%

91% ACCURACY

Precision: Correctly detected drowsy / Total predicted drowsy
Recall: Correctly detected drowsy / Total actual drowsy
Chart

High precision & recall confirm system reliability for real-time drowsiness detection.

Made byBobr AI
18

LIMITATIONS

College Logo
Warning

Poor Lighting Conditions

Performance degrades in very low-light or high-glare environments. Infrared cameras needed for night use.

Face Detection Dependency

System requires clear face visibility. Accuracy drops with masks, head turns, or occlusions.

Fixed EAR Threshold

Single threshold (0.21) may not suit all users due to variations in eye size and structure.

Single Feature Reliance

Only EAR used; head nod, yawning, and other fatigue signs not detected.

No Hardware Integration

Currently software-only; not yet embedded in vehicle systems.

Rule-Based Approach

Simple threshold logic; no adaptive learning to individual user patterns.

These limitations guide the path toward future improvements and advanced ML integration.

Made byBobr AI
19

CONCLUSION

College Logo
Check

Successfully developed a real-time Driver Drowsiness Detection System

Check

Eye Aspect Ratio (EAR) proved effective as the primary detection feature

Check

Mediapipe Face Mesh provides accurate, fast facial landmark detection

Check

Lightweight ML approach achieves ~91% accuracy without GPU

Check

Instant audio alert via Pygame ensures timely driver notification

Check

System operates on standard CPU hardware — cost-effective & accessible

"This project demonstrates that Computer Vision and AI can significantly enhance road safety by detecting driver fatigue in real time."

System Specifications

Accuracy 91%
Performance 20–30 FPS
Hardware Standard CPU
Input Webcam
Alert System Audio (Pygame)
Core Library Mediapipe + OpenCV
Made byBobr AI

20

FUTURE SCOPE

College Logo
🧠

Deep Learning Integration

Implement CNN models to analyze facial images directly for improved accuracy under challenging conditions.

📊

Large-Scale Dataset Training

Collect diverse user datasets for training adaptive ML models.

🚗

Vehicle Integration

Embed system into vehicle dashboards with OBD integration and real hardware alerts.

🌙

Infrared / Night Vision

Add infrared camera support for reliable performance in low-light and nighttime conditions.

📱

Mobile Application

Develop portable app for smartphones to make system widely accessible.

🎯

Multi-Feature Detection

Include head nodding, yawning detection, and heart rate monitoring for comprehensive fatigue analysis.

Rule-Based → SVM/Logistic Regression → CNN → Multi-Modal Deep Learning System

Made byBobr AI
21

REFERENCES

[1]
Soukupova & Cech (2016) "Real-Time Eye Blink Detection using Facial Landmarks" CVWW 2016
[2]
Viola & Jones (2001) "Rapid Object Detection using Boosted Cascade of Simple Features" CVPR 2001
[3]
Lugaresi et al. (2019) "MediaPipe: A Framework for Building Perception Pipelines" Google Research
[4]
Bradski (2000) "The OpenCV Library" Dr. Dobb's Journal of Software Tools
[5]
Ngxande et al. (2017) "Driver Drowsiness Detection Using Behavioral Measures" IEEE ROBMECH 2017
[6]
Zhang et al. (2020) "Drowsiness Detection with EAR using CNN" IEEE Access
Logo Right

THANK YOU

We extend our sincere gratitude to our guide and faculty for their support.

Baharul Islam | Bhaskar Sha | Bongso Hanse | Hafizur Rahman | Md. Dilshan Ul Hassan Choudhury

Under the guidance of:

Mr. Swmdwm Basumatary, Asst. Professor, CSE

Barak Valley Engineering College, Assam — May 2026

College Logo
Made byBobr AI
Bobr AI

DESIGNER-MADE
PRESENTATION,
GENERATED FROM
YOUR PROMPT

Create your own professional slide deck with real images, data charts, and unique design in under a minute.

Generate For Free

Driver Drowsiness Detection using Computer Vision & ML

Learn how to build a real-time driver drowsiness detection system using Mediapipe, Eye Aspect Ratio (EAR), and Python for improved road safety.

A Mini Project Report

DRIVER DROWSINESS DETECTION

USING COMPUTER VISION

Department of Computer Science & Engineering

Barak Valley Engineering College

(Affiliated to ASTU, Approved by AICTE)

Nirala, Sribhumi, Assam - 788701

May 2026

STATUS: ACTIVE

Baharul Islam | Bhaskar Sha | Bongso Hanse | Hafizur Rahman | Md. Dilshan Ul Hassan Choudhury

TABLE OF CONTENTS

02

Introduction

Need of Machine Learning

Literature Survey

Existing Technology

Problem Statement

Objectives

Requirement Analysis

Data Flow Diagram

Use Case Diagram

Methodology

Eye Aspect Ratio (EAR)

Classification Model

Dataset Description

Experimental Setup

Results & Discussion

Performance Evaluation

Limitations

Conclusion

Future Scope

References

03

INTRODUCTION

Road accidents are a major global concern — driver drowsiness is a leading cause.

Fatigue, lack of sleep, and long driving hours reduce alertness and reaction time.

Traditional vehicle safety systems monitor external conditions only (braking, collision avoidance) — NOT the driver's internal state.

Need for intelligent real-time systems to detect early signs of drowsiness.

This project proposes a Driver Drowsiness Detection System using Computer Vision and Machine Learning.

Core technique: Eye Aspect Ratio (EAR) computed via Mediapipe Face Mesh facial landmarks.

Lightweight, real-time, cost-effective — no GPU required.

~20%

of fatal accidents caused by drowsy driving

04

NEED OF MACHINE LEARNING

Feature Extraction

EAR converts visual data into numerical form for analysis

Binary Classification

Classifies driver as Alert (0) or Drowsy (1)

Decision Boundary

EAR < 0.21 threshold acts as ML decision boundary

Temporal Analysis

Frame count distinguishes blinking from drowsiness

Scalability

Extendable to SVM, Logistic Regression, CNN models

ML enables adaptability across different users, lighting, and driving conditions.

05

LITERATURE SURVEY

Paper 1

Real-Time Eye Blink Detection using Facial Landmarks

Introduces EAR concept. EAR remains constant when eyes open, drops when closed. Efficient for real-time use.

Paper 2

Driver Drowsiness Detection Using Computer Vision

Uses OpenCV webcam-based eye tracking. Non-intrusive, cost-effective.

Paper 3

Drowsy Driver Detection Using ML

SVM and neural networks for classification. High accuracy but needs large datasets.

Paper 4

Real-Time Fatigue Detection Using Mediapipe

Mediapipe Face Mesh for accurate landmarks. Faster than Haar cascades.

Paper 5

Vision-Based Driver Monitoring System

Combines eye detection + head movement. Multi-feature reduces false detection.

06

EXISTING TECHNOLOGY

OpenCV

Open Source Computer Vision Library

Lightweight • Cross-platform • Python-ready

Mediapipe Face Mesh

Google's Real-Time Perception Framework

Accurate • Fast • Cross-platform

ML Techniques

Classification & Feature Analysis

Scalable • Adaptable • Efficient

07

PROBLEM STATEMENT

Most vehicle safety systems monitor roads, NOT the driver.

Lack of Real-Time Monitoring

Traditional systems don't continuously track driver alertness.

Inability to Detect Early Signs

Subtle behavioral changes like slow blinking go undetected.

High Cost of Advanced Systems

Infrared/EEG-based solutions are expensive, limited to premium vehicles.

Dependency on Manual Observation

Drivers may not recognize their own fatigue.

Limited Adaptability & Accuracy

Fixed rules fail across users, lighting, and environments.

No Lightweight Real-Time Solutions

Deep learning models need powerful hardware, impractical for real-time use.

This project addresses all the above challenges with a lightweight, real-time, cost-effective CV-based solution.

08

OBJECTIVES OF THE PROJECT

01 — Real-Time Detection

Develop a real-time drowsiness detection system using webcam input.

02 — ML Classification

Apply Machine Learning concepts for binary classification (Alert / Drowsy).

03 — Eye Closure via EAR

Detect eye closure using Eye Aspect Ratio (EAR) feature extraction.

04 — Instant Audio Alert

Trigger immediate sound alarm (Pygame) when drowsiness is detected.

05 — Low-Cost Solution

Create a lightweight system deployable on standard CPU hardware.

Under the Guidance of: Mr. Swmdwm Basumatary, Asst. Professor, Dept. of CSE, BVEC

10

DATA FLOW DIAGRAM (DFD)

Level 0 — Context Diagram

Driver

Facial video (webcam)

Drowsiness Detection

System

Drowsiness alert

Alert Output

Level 1 — Detailed Flow

Driver

— Live video stream

P1:

Webcam Input & Frame Capture

— Continuous frame extraction

P2:

Face Detection (Mediapipe)

— Facial landmark localization

P3:

Eye Landmark Extraction

— 6 eye region points per eye

P4:

EAR Calculation

— Eye openness ratio

P5:

ML Classification

— Drowsy vs Alert decision

Alert Trigger

— Alarm/buzzer

Normal State

— Continue monitoring

USE CASE DIAGRAM

DETAILED METHODOLOGY

Data Acquisition

Webcam captures live video

Frame-by-frame real-time extraction

Non-intrusive, continuous

Feature Extraction

Mediapipe detects 468 facial landmarks

Eye region coordinates extracted

EAR computed (6 points per eye)

Feature Processing

EAR compared to threshold (0.21)

Frame counter incremented/reset

Noise filtered (blink vs drowsiness)

Classification

Rule-based ML binary classifier

EAR &lt; 0.21 AND frames &gt; 30 &rarr; Drowsy

Otherwise &rarr; Alert

Output & Alert

Audio alarm via Pygame

Visual warning on screen

System continues monitoring

EYE ASPECT RATIO (EAR)

What is EAR?

EAR measures eye openness using 6 facial landmark points per eye.

EAR Threshold Visualization

CLASSIFICATION MODEL

Rule-Based ML Classifier

Binary classification using EAR + frame count as features

DROWSY

ALERT

Binary Classes

0 = ALERT

— Eyes open, normal

1 = DROWSY

— Prolonged eye closure

Temporal Analysis

Normal blink: EAR drops briefly → counter resets

Drowsiness: EAR stays low > 30 frames → alert triggered

Frame counter prevents false positives

EAR Value Computed

Increment Frame Counter

Reset Frame Counter

TRIGGER ALERT

Drowsy

Continue Monitoring

15

DATASET & EXPERIMENTAL SETUP

Dataset Description

No predefined dataset used — real-time dynamic data from webcam.

Primary Feature: EAR (Eye Aspect Ratio) computed per frame

Frame No.

EAR Value

Frame Count

State

Data generated dynamically — no storage needed → lightweight system

Experimental Setup

Python 3.x — Core Programming Language

OpenCV — Video capture & image processing

Mediapipe — Face mesh landmark detection

NumPy — Numerical computations

Pygame — Audio alert system

Standard CPU — No GPU required

Webcam — Real-time video input

20–30 FPS processing speed

16

RESULTS & DISCUSSION

Case 1: Alert State

Eyes Fully Open

0.32

Frame Count: 0

✅ GREEN ALERT — No action taken

System continues monitoring

Case 2: Normal Blink

Brief Eye Closure

0.18

Frame Count: 3 (low)

🟡 AMBER TRANSITION — No false alert

Counter resets — blink detected correctly

Case 3: Drowsy State

Prolonged Eye Closure

0.16

Frame Count: 32+

🔴 RED DROWSY — Alert Triggered!

Audio alarm activated immediately

System accurately distinguishes blinking from drowsiness with real-time alert response.

17

PERFORMANCE EVALUATION

Confusion Matrix

Accuracy = (TP + TN) / Total = (45+46)/100 ≈ 91%

91% ACCURACY

High precision & recall confirm system reliability for real-time drowsiness detection.

18

LIMITATIONS

Poor Lighting Conditions

Performance degrades in very low-light or high-glare environments. Infrared cameras needed for night use.

Face Detection Dependency

System requires clear face visibility. Accuracy drops with masks, head turns, or occlusions.

Fixed EAR Threshold

Single threshold (0.21) may not suit all users due to variations in eye size and structure.

Single Feature Reliance

Only EAR used; head nod, yawning, and other fatigue signs not detected.

No Hardware Integration

Currently software-only; not yet embedded in vehicle systems.

Rule-Based Approach

Simple threshold logic; no adaptive learning to individual user patterns.

These limitations guide the path toward future improvements and advanced ML integration.

19

CONCLUSION

Successfully developed a real-time Driver Drowsiness Detection System

Eye Aspect Ratio (EAR) proved effective as the primary detection feature

Mediapipe Face Mesh provides accurate, fast facial landmark detection

Lightweight ML approach achieves ~91% accuracy without GPU

Instant audio alert via Pygame ensures timely driver notification

System operates on standard CPU hardware — cost-effective & accessible

This project demonstrates that Computer Vision and AI can significantly enhance road safety by detecting driver fatigue in real time.

91%

20–30 FPS

Standard CPU

Webcam

Audio (Pygame)

Mediapipe + OpenCV

FUTURE SCOPE

Deep Learning Integration

Implement CNN models to analyze facial images directly for improved accuracy under challenging conditions.

Large-Scale Dataset Training

Collect diverse user datasets for training adaptive ML models.

Vehicle Integration

Embed system into vehicle dashboards with OBD integration and real hardware alerts.

Infrared / Night Vision

Add infrared camera support for reliable performance in low-light and nighttime conditions.

Mobile Application

Develop portable app for smartphones to make system widely accessible.

Multi-Feature Detection

Include head nodding, yawning detection, and heart rate monitoring for comprehensive fatigue analysis.

Rule-Based → SVM/Logistic Regression → CNN → Multi-Modal Deep Learning System

21

REFERENCES

[1]

Soukupova & Cech (2016)

Real-Time Eye Blink Detection using Facial Landmarks

CVWW 2016

[2]

Viola & Jones (2001)

Rapid Object Detection using Boosted Cascade of Simple Features

CVPR 2001

[3]

Lugaresi et al. (2019)

MediaPipe: A Framework for Building Perception Pipelines

Google Research

[4]

Bradski (2000)

The OpenCV Library

Dr. Dobb's Journal of Software Tools

[5]

Ngxande et al. (2017)

Driver Drowsiness Detection Using Behavioral Measures

IEEE ROBMECH 2017

[6]

Zhang et al. (2020)

Drowsiness Detection with EAR using CNN

IEEE Access

We extend our sincere gratitude to our guide and faculty for their support.

Baharul Islam | Bhaskar Sha | Bongso Hanse | Hafizur Rahman | Md. Dilshan Ul Hassan Choudhury

Under the guidance of:

Mr. Swmdwm Basumatary, Asst. Professor, CSE

Barak Valley Engineering College, Assam — May 2026