Driver Drowsiness Detector

Overview

The Driver Drowsiness Detector is an AI-powered system designed to reduce road accidents caused by driver fatigue. By analyzing facial landmarks and eye aspect ratios using a webcam feed, the application can accurately determine when a driver is becoming drowsy and trigger visual and audible alerts.

The problem

Drowsy driving is a leading cause of traffic accidents, often resulting from drivers unintentionally falling asleep or losing focus at the wheel. There is a need for an automated, real-time solution to monitor driver alertness and prevent accidents.

The solution

This system leverages deep learning and computer vision to continuously monitor a driver's facial features and eye states. When drowsiness or microsleeps are detected, the application issues immediate alerts, helping to keep drivers attentive and safe.

Features

Real-Time Drowsiness Detection

Monitors driver's eyes and facial features in real time to detect drowsiness or microsleeps.

Eye Aspect Ratio Calculation

Calculates EAR from facial landmarks to determine if eyes are closed for extended periods.

Deep Learning Classification

Uses a trained CNN to distinguish between open and closed eye states, improving detection accuracy.

Immediate Alerts

Triggers visual and audible alarms when drowsiness is detected, warning the driver instantly.

Configurable Thresholds

Allows customization of EAR and time thresholds to adapt to different drivers and environments.

Architecture

The application uses OpenCV to capture and process video frames, MediaPipe for facial landmark detection, and a pre-trained CNN to classify eye states (open or closed). The logic computes the Eye Aspect Ratio (EAR) and triggers alerts when drowsiness is detected. The system is designed for real-time processing and can be deployed on laptops or embedded systems with a webcam.

Video Processing

OpenCV
Webcam Capture
Frame Preprocessing

Facial Landmark Detection

MediaPipe
68-point Landmark Model

Deep Learning Model

CNN
Keras
TensorFlow
Eye State Classification

Alert System

Visual Alerts
Audio Alerts
Threshold Logic

Screenshots

Problem Description — Overview of the driver drowsiness detection problem and system objectives.

Objective — Project objective and classification goals for detecting driver fatigue.

Detection Approaches — Comparison of Machine Learning and Deep Learning detection approaches.

Implemented Models — Overview of the implemented Logistic Regression, Random Forest, CNN, and Vision Transformer models.

Workflow — Complete workflow from preprocessing to driver state prediction.

Machine Learning Models — Machine Learning pipeline using EAR, MAR, and PERCLOS features.

Deep Learning Models — Deep Learning pipeline using CNN and Vision Transformer architectures.

CNN Model Visualization — Visualization of the Convolutional Neural Network architecture used for eye image classification.

Vision Transformer Visualization — Visualization of the Vision Transformer architecture for eye-state classification.

Machine Learning Results — Evaluation results for Logistic Regression and Random Forest models.

CNN Results — CNN confusion matrix, classification report, and training performance.

Vision Transformer Results — Vision Transformer evaluation metrics, confusion matrix, and learning curves.

Challenges

Real-Time Performance

Ensuring low-latency video processing and accurate detection required optimization of both the computer vision pipeline and neural network inference.

Robustness to Lighting & Angles

The system needed to work reliably across varying lighting conditions and different driver positions, requiring data augmentation and careful calibration.

Lessons learned

Combining classical computer vision with deep learning improves reliability for real-world applications.

Real-time systems must balance accuracy with speed, especially for safety-critical use cases.

Customizable parameters are essential for adapting AI solutions to diverse environments and users.

Next project

ExpoAlAlam