CNN-Project---Driver-Drowsiness-Detection

This project is a Drowsiness Detection System that uses a Convolutional Neural Network (CNN) and OpenCV to monitor a user's eyes through a webcam and detect signs of drowsiness in real-time.

Project Description

The system works by:

• Capturing video frames using a webcam via OpenCV
• Detecting the eyes using Haar cascades
• Classifying whether eyes are open or closed using a trained CNN model
• Raising an alert (e.g., sound) if drowsiness is detected over consecutive frames

File structure

• support_files/

  • Alert.mp3
  • my_right_eye.jpg
  • my_left_eye.jpg
  • closed_eye.jpg
  • Graphs for README

• Driver_Drowsiness_integrated.py -> Main file to run the application

• Driver_Drowsiness_CNN.ipynb -> File to train a CNN model

• Driver_Drowsiness_MobileNetV2.ipynb -> File implementing the pretrained model

• image_data_processing.ipynb -> File to preprocess your trainig data

CNN Model

The Convolutional Neural Network (implemented in Driver_Drowsiness_CNN.ipynb) is trained to classify eye states:

• Open
• Closed

It includes:

• Convolutional and MaxPooling layers
• Dropout for regularization
• Dense output layer for binary classification

It achieved an accuracy of around 96% on the test data

Pretrained model MobileNetV2

I have used a pretrained model to acchieve higher accuracy for the same application. It reduces the training time and works on RGB images.

• Implemented in Driver_Drowsiness_MobileNetV2.ipynb
• To use this file make sure to turn on flag_rgb in the image_data_processing.ipynb file.
• It achieved an accuracy of around 99.8% on the test data

Drowsiness Detection Logic

Implemented in Driver_Drowsiness_integrated.ipynb:

• Streams webcam frames via OpenCV
• Detects eye regions using Haar cascade classifier