🚗 Edge Driver Drowsiness & Distraction Alert System

A lightweight, real-time Computer Vision safety pipeline engineered for edge devices.
Monitors driver vigilance, micro-sleeps, yawning, head gaze orientation, and mobile phone usage with sub-second latency and zero frame lag.

---

---

📌 Overview

Driver fatigue and distraction are leading contributors to road collisions worldwide. This system provides a non-intrusive, vision-based Driver Monitoring System (DMS) that evaluates driver cognitive alertness continuously in real-time.

By combining facial geometric landmarks with an optimized edge object detector, the system detects acute events (such as sudden eye closures or phone handling) while tracking cumulative, progressive drowsiness over a rolling 60-second temporal window.

---

⚡ Core Features

• ⚡ Ultra-Fast Micro-Sleep Detection: Eye Aspect Ratio (EAR) persistence threshold tuned to trigger within ~0.35 seconds while filtering natural ocular blinks.
• 🥱 Oral Fatigue & Yawn Frequency: Tracks Mouth Aspect Ratio (MAR) with edge-triggered counters to prevent multiple counts for a single yawning event.
• 🧭 3D Perspective-n-Point Head Pose: Computes head Yaw, Pitch, and Roll using canonical 3D facial geometry to flag glances away from the road.
• 📱 Edge Object Detection (ONNX): Quantized YOLOv8n running via onnxruntime CPU with strict confidence and bounding-box aspect filters to eliminate hand false-positives.
• 📈 Predictive Fatigue Engine: 60-second rolling sliding window aggregating PERCLOS (Percentage of Eye Closure), yawn spikes, and gaze deviations into a 0–100% Fatigue Index.
• 🖥️ Dynamic Adaptive HUD: Real-time vector graphics downscaled via cv2.INTER_AREA interpolation for sharp, legible text across any window resolution or aspect ratio.
• 🔊 Asynchronous Audio Warnings: Stereo tone synthesized on a dedicated worker thread via Pygame to guarantee uninterrupted video capture throughput.

---

🛠️ System Pipeline Architecture

               Webcam Stream (1280x720 Native Widescreen)
                                   │
                                   ├──> [Frame Preprocessing & Vector HUD]
                                   │
           ┌───────────────────────┴────────────────────────┐
           │                                                │
           ▼ (Every Frame)                                  ▼ (Every 2nd Frame)
MediaPipe FaceMesh (468 Points)                   YOLOv8n ONNX Engine
  ├── Eye Aspect Ratio (EAR)                        └── Cell Phone Detection
  ├── Mouth Aspect Ratio (MAR)                            (Aspect Filtered)
  └── 3D solvePnP (Yaw / Pitch)                             │
           │                                                │
           └───────────────────────┬────────────────────────┘
                                   │
                                   ▼
              Temporal Fatigue Engine (Rolling 60s Buffer)
               ├── PERCLOS % Accumulation
               ├── Yawn Frequency Tracker
               └── Inattention / Gaze Penalty
                                   │
                                   ▼
                     Decision Logic & Telemetry Dock
               ├── Asynchronous Audio Alerts (Pygame)
               └── Proportional Anti-Aliased HUD

---

📊 Detection Logic & Calibrated Thresholds

Metric	Monitored Behavior	Trigger Threshold	Alert State
EAR	Acute Micro-Sleep / Closed Eyes	EAR < 0.20 sustained for >= 9 frames (~0.35s)	DROWSINESS DETECTED
MAR	Yawning & Drowsiness Spikes	MAR > 0.65 sustained for >= 12 frames	YAWNING
Head Pose	Gaze Distraction / Looking Away	|Yaw| > 25° or |Pitch| > 20°	DISTRACTED GAZE
Object Detection	Mobile Phone Handling	Class 67, Confidence >= 0.50, Aspect Filtered	PHONE USE
Fatigue Index	Cumulative Fatigue (Rolling 60s)	Score >= 60% (PERCLOS + Yawn Count + Inattention)	FATIGUE WARNING

---

📂 Project Structure

Driver-Drowsiness-Distraction-AlertEdge/
├── core/
│   ├── eye_detector.py        # EAR computation & eye state tracking
│   ├── mouth_detector.py      # MAR computation & yawn detection
│   ├── head_pose.py           # 3D solvePnP pose estimation (Yaw/Pitch/Roll)
│   ├── object_detector.py     # YOLOv8n ONNX inference & geometry filter
│   └── fatigue_scorer.py      # Rolling 60s temporal PERCLOS & fatigue engine
├── models/
│   └── yolov8n.onnx           # Lightweight ONNX detection model (~12.3 MB)
├── utils/
│   └── sound_alert.py         # Multi-threaded audio tone generator
├── requirements.txt           # Pinned production dependencies
├── main.py                    # Main pipeline, arbitration & vector HUD
├── run.bat                    # One-click Windows runner
└── README.md                  # Project documentation

---

🚀 Getting Started

Prerequisites

• Python 3.10 or 3.11
• Integrated or external USB Webcam
• Windows 10/11 (or Linux/macOS)

---

Step-by-Step Setup

1. Clone the Repository

    git clone https://github.com/AreebaAminn/Driver-Drowsiness-Distraction-AlertEdge.git
    cd Driver-Drowsiness-Distraction-AlertEdge

2. Set Up a Virtual Environment

   python -m venv venv
   .\venv\Scripts\Activate.ps1

3. Install Dependencies

   pip install -r requirements.txt

---

💻 Running the Monitor

• Option A: Command Line

  python main.py

• Option B: One-Click Execution (Windows)
  Double-click run.bat in the project root.

Key Controls: Press q while focused on the display window to cleanly release camera resources and terminate the session.

---

⚖️ License

Distributed under the MIT License. See LICENSE for more information.