Blurring Faces in Videos

An approach to blur faces in videos using Deep Learning.

Table of Contents

1. Introduction
2. Setup
3. How to Run
4. How Does It Work?
5. Contributions

1. Introduction

It is a common practice for faces of people to be blurred in videos to preserve privacy of individuals. There exists a plethora of software that can be used to blur human faces in a video file. This project is an attempt to blur faces using Deep Learning techniques.

This project aims to:

• Detect and blur faces that appear in videos with minimum manual effort.

• Let users to choose specific faces to be blurred using face recognition techniques [Work In Progress!!]

Below is a demonstration of how this project works. Left side is the input video (of Sebastian Vettel, from Formula 1 Official YouTube Channel) and the right side is the output taken from the face blurring script.

2. Setup

The requirements have to be installed in order to run scripts on a local computer. It is recommended to create a Python virtual environment to run the scripts of this repository and to keep track of requirements.

First, clone this repository to the computer.

git clone https://github.com/theSLWayne/Video_Face_Blur.git

Then, create a Python virtual environment.

python -m venv facial_blur

Activate the created virtual environment.

• For Linux:

  source facial_blur/bin/activate
  

• For Windows:

  facial_blur\Source\activate
  

Install dependencies to the virtual environment.

pip install -r requirements.txt

After everything is installed, all scripts of the repository can be run inside the virtual environment.

In order to deactivate the virtual environment, simply use deactivate command.

3. How to Run

blur_faces.py can be used to blur faces in a video. The syntax as below:

python blur_faces.py --video_path path/to/the/video/file
        --output_directory folder/path/to/save/output/video
        --show_frames y

Arguments

Argument	Description	Notes
--video_path / -p	Path of the input video	Must be a valid path for a video file.
--output_directory / -od	Path of the folder where the output video should be saved to	Not mandatory. The video will be saved to ./outputs folder when no output path is provided.
--show_frames / -s	Whether to display frames as they are processed	Not Mandatory. Defaulted to False.

• If the provided output path does not exist, the output video will be saved to the default location (./outputs).

4. How does it work?

The workflow of face blurring can be broken down to a few key steps. Below is a higher level overview of the workflow.

• Audio, FPS (frames per second), and the list of frames are extracted from the input video.

• Each frame is sent through the OpenCV face detection model to identify faces.

• If there are faces identified, they are blurred using Gaussian Blurring and then the frame is appended to a separate frames list. If no faces were detected in a frame, it is appended to the frames list without further processing.

• Updated frames list is converted into a video using the FPS count of the original video.

• Audio extracted from the original video is added to the video created using updated frames.

• Final video is written into the output directory.

5. Contributions

Feel free to contribute to the project. Here are some of the suggested improvements:

• Improve the face detection model to detect side faces.

• Add face recognition to let users select faces that should be blurred in a video.

↑ Top