Real-Time Face De-Identification and Inside-Outside Classification

This repository contains two Python scripts and two pre-trained .h5 models for real-time video processing:

1. real-time-de-identification.py

   • Identifies faces in real-time from a video feed or video file.
   • Blurs or de-identifies recognized faces using various methods.

2. outside_remover.py

   • Classifies frames from a video as inside or outside using a pre-trained model.
   • Saves frames classified as inside to a new video file.

Requirements

Python Dependencies

The scripts require the following libraries:

• opencv-python
• numpy
• tensorflow
• scipy

Install dependencies using pip:

pip install opencv-python numpy tensorflow scipy

Model Files

• face_classifier.h5: Used in real-time-de-identification.py to classify faces as me or not_me.
• inside_outside.h5: Used in outside_remover.py to classify video frames as inside or outside.

Place the .h5 files in the following directories:

• face_classifier.h5 in models/
• inside_outside.h5 in Documents/5th_sem/Project/

Usage

1. Real-Time Face De-Identification

Run the real-time-de-identification.py script to process a live video feed or a saved video file.

Configuration

• Set VIDEO_FILE to the path of your video file if not using a live feed.
• Ensure face_classifier.h5 is in the models/ directory.

Run

python real-time-de-identification.py

Features

• Identifies faces using OpenCV's Haar Cascade.
• Uses a pre-trained classifier to determine if a face matches me.
• Blurs or de-identifies recognized faces using:
  • Gaussian blur
  • Pixelation (mosaic effect)

2. Inside-Outside Frame Classification

Run the outside_remover.py script to process a video file and save only inside frames to a new video file.

Configuration

• Set video to the path of your input video file.
• Set output to the desired path for the processed video.
• Ensure inside_outside.h5 is in the Documents/5th_sem/Project/ directory.

Run

python outside_remover.py

Features

• Classifies frames as inside or outside using a pre-trained model.
• Saves frames classified as inside to a new video file.

File Structure

project/
├── models/
│   └── face_classifier.h5
├── Documents/
│   └── 5th_sem/
│       └── Project/
│           └── inside_outside.h5
├── real-time-de-identification.py
├── outside_remover.py
└── README.md

Notes

• Ensure your Python environment has all the required dependencies.
• Modify file paths in the scripts to match your local directory structure if needed.

License

This project is licensed under the MIT License. See the LICENSE file for details.

Author

Praharsh Gurudatta