Drowsy Driver Detection System

This is a fork of varshathennarasu/DrowsyDriverDetection. It was a 5-person team project built through the AI Student Collective at UC Davis during the Winter 2025 cycle, where it won Best Execution. I worked on it as part of that team.

Overview

This is a real-time system that detects signs of driver drowsiness from a live webcam feed to help prevent fatigue-related accidents. It analyzes facial landmarks frame by frame and pairs threshold-based eye tracking with a machine learning model trained on a labeled Kaggle dataset, then triggers an audio alert when it detects fatigue.

Kaggle Dataset: https://www.kaggle.com/datasets/prasadvpatil/mrl-dataset

Features

• Real-Time Eye Blink Detection: Monitors eye aspect ratio (EAR) to detect signs of drowsiness.
• Facial Landmark Detection: Identifies facial regions to track eye and mouth movements using a pre-trained model.
• Machine Learning Integration: Trains and evaluates a model using a Kaggle dataset for more accurate drowsiness and intoxication detection.
• Alert Mechanism: Plays a warning sound if drowsiness or intoxication is detected.
• Efficient Processing: Optimized for quick response times using imutils, scipy, and OpenCV.

Technologies Used

• Python: Main programming language.
• OpenCV: For video capture and facial detection.
• dlib: For detecting facial landmarks.
• Pygame: For audio alerts.
• scipy: To compute Euclidean distances for EAR.
• Pandas & NumPy: For data manipulation during model training.
• Scikit-learn: For training and evaluating the machine learning model.

Kaggle Dataset

This project utilizes a dataset from Kaggle containing labeled images of drowsy and alert drivers. The dataset was preprocessed and split into training and testing sets to develop a robust machine learning model capable of detecting drowsy behavior.

Project Structure

.
├── data/                                       # Dataset folder
│   ├── train/                                 # Training data
│   └── test/                                  # Testing data
├── shape_predictor_68_face_landmarks.dat      # Pre-trained model for facial landmarks
├── music.wav                                  # Alert sound file
├── drunk_drowsy_detection.py                  # Main script
└── README.md                                  # Project documentation

How It Works

1. Eye Aspect Ratio (EAR): Calculates EAR from facial landmarks. If EAR falls below a threshold, the driver is likely drowsy.
2. Facial Landmark Analysis: Tracks mouth movements to detect potential yawning (a sign of drowsiness).
3. Machine Learning Model: Analyzes features extracted from facial landmarks and predicts driver status (alert or drowsy).
4. Real-Time Alerts: When thresholds are breached or the model predicts drowsiness, the system triggers an alert.

Getting Started

Prerequisites

Ensure you have the following Python libraries installed:

• OpenCV
• dlib
• pygame
• imutils
• scipy
• pandas
• numpy
• scikit-learn

Install the required packages using:

pip install opencv-python dlib pygame imutils scipy pandas numpy scikit-learn

Usage

1. Download the shape predictor model and place it in the project directory.
2. Ensure the Kaggle dataset is downloaded and placed in the data/ directory.
3. Run the script:

   python drunk_drowsy_detection.py

4. The system will start the camera, monitor driver behavior in real-time, and utilize the trained model for predictions.

Configuration

• Modify thresh (default: 0.25) in the script to adjust the sensitivity of EAR.
• Adjust frame_check (default: 20) to set the number of consecutive frames required to trigger the alert.

Future Improvements

• Integrating additional features like head pose estimation for better accuracy.
• Deploying the system on mobile devices or in-vehicle hardware.
• Incorporating deep learning for advanced behavioral analysis.

Acknowledgments

This project uses:

• The shape_predictor_68_face_landmarks model from dlib.
• A labeled dataset from Kaggle(https://www.kaggle.com/datasets/prasadvpatil/mrl-dataset) for training and testing.