driver_drowsiness_detection

Driver Drowsiness Detection System

An Edge-AI Safety System for Real-Time Fatigue Monitoring

📌 Project Overview

This project is an advanced real-time computer vision framework designed to mitigate highway transit accidents caused by operator exhaustion and micro-sleep events [AnshumanSrivastava108/Real-Time-Drowsiness-Detection-System].

By processing live camera frames, the system maps facial telemetry landmarks using spatial equations to calculate eye and mouth ratios, instantly executing audio alarms when safety limits are breached.

🚀 Key System Features

Dynamic HUD Status Display: Real-time driver conditions are flashed directly onto the video feed layout (AWAKE, DROWSY (WARNING), or DROWSY (CRITICAL!)).
Unified Chrome-Green Mesh: Standardized bright green boundaries (BGR: 0, 255, 0) draw clean outlines around ocular and labial zones for optimal low-light tracking [AnshumanSrivastava108/Real-Time-Drowsiness-Detection-System].
Multi-Stage Escalated Alarm:
- Soft Warning: Sounds single tones during repetitive yawn loops.
- Critical Siren: Sounds aggressive, rapid double-beeps during deep micro-sleeps.
Threaded Audio Subprocesses: Uses background threads to play alert buzzers asynchronously, preventing display frame drops or screen lag [AnshumanSrivastava108/Real-Time-Drowsiness-Detection-System].
Black Box Event Logging: Automatically records event logs with calendar timestamps into a local repository file for administrative safety audits.

📁 Repository Directory Map

driver_drowsiness_detection/
│
├── models/
│   └── shape_predictor_68_face_landmarks.dat  # Neural network weights database
│
├── utils/
│   ├── __init__.py
│   └── metrics.py                             # Formulas file (EAR & MAR math)
│
├── config.py                                  # Calibration thresholds hub
├── main.py                                    # Central camera loops engine
├── requirements.txt                           # Software dependencies list
└── drowsiness_logs.txt                        # Automated telemetry data file

🔬 Mathematical Framework & Diagnostics

The core analytical loops rely on scaling-invariant mathematical fractions, guaranteeing accurate processing regardless of how close or far the driver sits from the lens.

1. Eye Aspect Ratio (EAR)

Calculates horizontal and vertical eyelids vectors: [\text{EAR} = \frac{\vert{}\vert{}p_2 - p_6\vert{}\vert{} + \vert{}\vert{}p_3 - p_5\vert{}\vert{}}{2 \vert{}\vert{}p_1 - p_4\vert{}\vert{}}]

🟢 STATUS: AWAKE → EAR ≥ 0.25
🟠 STATUS: DROWSY (WARNING) → EAR < 0.25 (Triggers soft loop after 10 frames)
🔴 STATUS: DROWSY (CRITICAL!) → EAR < 0.25 (Triggers double-beep sirens after 20 frames)

2. Mouth Aspect Ratio (MAR)

Calculates inner lip dilation layouts during active driving shifts: [\text{MAR} = \frac{\vert{}\vert{}p_{51} - p_{59}\vert{}\vert{} + \vert{}\vert{}p_{53} - p_{57}\vert{}\vert{}}{2 \vert{}\vert{}p_{49} - p_{55}\vert{}\vert{}}]

Normal Baseline → MAR ≤ 0.60
🟠 STATUS: YAWNING DETECTED → MAR > 0.60 (Triggers cooldown logging after 15 frames)

📊 Black Box Logs Sample (`drowsiness_logs.txt`)

The system appends clean chronological records automatically during active run states:

[2026-07-21 14:30:05] SYSTEM START: Safety monitoring initialized.
[2026-07-21 14:32:14] ALERT: Driver was Drowsy - Duration: 2.45 seconds
[2026-07-21 14:35:40] NOTICE: Yawn Pattern Logged
[2026-07-21 14:40:12] SYSTEM SHUTDOWN: Monitoring ended cleanly.

💻 Technical Setup & Execution

1. Dependencies Configuration

Deploy all mandatory dependencies into your localized sandbox environment by running:

pip install -r requirements.txt

2. Neural Weights Initialization

Download the pre-trained data file from the Official Dlib Container.
Save the compiled file as shape_predictor_68_face_landmarks.dat inside your models/ directory.

3. Running the Engine

Initialize the central driver dashboard panel loop using your terminal interface:

python main.py

🎮 Interface Controls

Key Command q: Safely stops camera input channels, flushes file writing streams, terminates background threads, and kills display windows cleanly.

👩‍💻 Author

MYLA DIVYA SAI SRI

Artificial Intelligence & Machine Learning (AI/ML) Student

LinkedIn: www.linkedin.com/in/myladivyasaisri
GitHub: https://github.com/myladivyasaisri