DrowseGuard v2 — Adaptive Driver Drowsiness Detection

Problem Statement

Driver drowsiness is a major road safety challenge in India, where long-distance travel, night driving, and fatigue contribute to preventable accidents on highways and urban roads. DrowseGuard addresses this issue with a real-time computer vision system that monitors eye closure and mouth opening patterns from a live webcam feed, detects drowsiness risk early, and provides immediate alerts to improve driver awareness and reduce crash probability.

Features

• Real-time EAR and MAR computation using MediaPipe facial landmarks
• Dual classifier inference with KNN and Gaussian Naive Bayes ensemble
• Audio alert subsystem with file-based alarm and generated fallback beep
• Multi-stage preprocessing pipeline visualization for interpretability
• Interactive Streamlit dashboard with live metrics and controls

Course Module Coverage

Module name	Specific functions
Module 1: Image Fundamentals and Enhancement	preprocessor.preprocess_frame(), preprocessor.apply_clahe(), preprocessor.detect_night_mode()
Module 2: Feature Extraction and Representation	feature_extractor.extract_hog_features(), feature_extractor.extract_combined_features()
Module 3: Segmentation, Edges, and Morphology	preprocessor.apply_canny_edges(), preprocessor.segment_eye_binary(), preprocessor.apply_morphological_ops()
Module 4: Statistical Inference and Temporal Metrics	perclos_engine.PerclosEngine.compute(), fatigue_scorer.FatigueScorer.compute(), eye_analyzer.EyeStateTracker.get_blink_rate_per_min()
Module 5: Detection and ML Integration	face_detector.FaceDetector.detect(), head_pose.HeadPoseEstimator.estimate(), classifier.EyeStateClassifier.fit() with sklearn.decomposition.PCA

Algorithms

PERCLOS is computed as the fraction of closed-eye frames in a rolling window:

$$ PERCLOS = \frac{\sum_{t=1}^{N} \mathbb{1}(EAR_t < 1.2 \cdot T_{EAR})}{N} $$

EAR is computed from six eye landmarks:

$$ EAR = \frac{|p_2 - p_6|_2 + |p_3 - p_5|_2}{2 \cdot |p_1 - p_4|_2} $$

Fatigue score is a weighted composite:

$$ S = 0.30S_{perclos} + 0.25S_{ear} + 0.15S_{blink} + 0.15S_{yawn} + 0.15S_{pose} $$

Installation

1. Clone the repository.

git clone https://github.com/your-username/drowseguard.git

2. Move into the project folder.

cd drowseguard

3. Install dependencies.

pip install -r requirements.txt

4. Run the application.

python main.py

Usage

Use the Live detection page for real-time inference, calibration, and alerts. Use the Session analytics page to inspect historical metrics, events, and snapshots from previous runs.

Session Data

Each run is stored under sessions/<session_id>/ with events.csv, metrics.csv, and a snapshots/ folder containing saved critical frames.

Tech Stack

Layer	Technology
Language	Python 3.10+
Computer Vision	OpenCV, MediaPipe
Numerical Computing	NumPy, SciPy
Machine Learning	scikit-learn, PCA
UI and Analytics	Streamlit, Plotly, pandas
Alerts and Persistence	pygame, CSV logging, JPEG snapshots

Author

Garv Anand