Computer Vision-Based Face Mask, Sunglasses, Caps, and Hoodies Detection

Project Overview

This project aims to develop a real-time system to detect if a human face is covered with masks, sunglasses, caps, or hoodies. The system is designed for applications in security monitoring, ATM security, and access control.

Technologies Used

• TensorFlow/Keras
• OpenCV
• FastAPI
• AWS S3
• Ngrok

Problem Statement

Unauthorized individuals covering their faces in restricted areas pose security concerns. This project implements a computer vision system to identify and alert when faces are covered by masks, sunglasses, caps, or hoodies.

Dataset Collection

Image Sources

• Internet
• Custom datasets

Categories

• Acceptable: Faces without any coverings.
• Alarming: Faces with masks, sunglasses, caps, or hoodies.

Preprocessing

• Image resizing to 224x224 pixels.
• Normalization.

Data Preparation

Data Download

Using Python scripts to download and categorize images.

Preprocessing

• Resizing images using OpenCV.
• Normalizing pixel values.

Data Storage

Organized in folders ('Acceptable' and 'Alarming') and uploaded to AWS S3 for access.

Model Development

Transfer Learning

• Base Model: ResNet50 pre-trained on ImageNet.
• Custom Layers: Added custom dense and output layers for binary classification.

Model Training

• Split dataset into training and testing sets.
• Used augmentation techniques for better generalization.

Model Training and Evaluation

Training

• Metrics: Accuracy, loss.
• Tools: TensorFlow/Keras.

Evaluation

• Validation on test set.
• Metrics: Test accuracy, confusion matrix, and classification report.

Model Deployment

Framework

FastAPI for serving the model.

Endpoints

• /start_prediction: Start video capture and prediction.
• /stop_prediction: Stop video capture.

Ngrok

Used to expose the local server to the internet.

Real-Time Prediction

Video Capture

Using OpenCV to access webcam and display predictions on video frames.

Prediction Logic

• Preprocessing each frame.
• Making predictions using the trained model.
• Displaying results on the video stream.

Challenges and Solutions

• Data Collection: Sourcing a diverse dataset.
• Model Accuracy: Improving accuracy through data augmentation and fine-tuning.
• Real-Time Processing: Optimizing prediction speed for real-time use.

Future Work

Improvements

• Increase dataset size and diversity.
• Enhance model accuracy and speed.

Additional Features

• Integrate with security systems.
• Develop a mobile application for on-the-go use.

Conclusion

This project successfully implements a computer vision system for detecting face coverings in real-time, enhancing security by providing timely alerts.

Contact Information

For any questions, feel free to reach out to me.