Intelligent Mirror Face Recognition

This repository contains code for integrating face recognition with the MagicMirror² framework. It supports running both locally on the RPI or it can read a video feed from an IP camera remotely and send results to the RPI.

Getting Started

Prerequisites

This module depends on quite a few python packages. Installing the following modules on a linux or Mac environment should be straight forward. However installing all packages for running the system locally on the RPI will take a lot of time and effort. I suggest googling how to install each individual package if you aim to run this on the RPI.

• Python 2.7
• macOS or Linux
• Face Recognition
• OpenCV
• Numpy
• Potentially more ...

Usage

This software requires interaction between various libraries for the system to work as intended.

• On the RPI:
  • MagicMirror²
  • MMM-Facial-Recognition-Custom
  • RPi-Cam-Web-Interface
• On remote machine:
  • This repository

Setup

The goal is to modify the MagicMirror² interface based on the recognized user. For this to work the system needs a coherent list of users both locally on the RPI and on the remote machine.

Adding users to the system

1. Create a folder in the /images directory for each person you want to recognize. The name is not important
   • Put as many pictures as you want of each person. At least more than 1 is suggested. The additional recognition time when using more images are near negligible. So feel free.
     • I suggest you precede each foldername by the index intended for that person. The directory is read in sorted order.
   • Set num_faces in conf.json to the number of images per person. Higher value = better accuracy
     • It cannot be lower than the number of images the folder with the least images has.
2. In the MagicMirror² configuration file config.js set the MMM-Facial-Recognition-2 module's users property to a list of the users name. Index 0 is reserved for unknown users. You can call the users whatever you want as long as the indices align with the order the /images folder is read. The strings entered in the list are used by MMM-facial-recognition-2 to swap modules based on current user.

Example:

• The /images directory contains the following folders:
  • 001_Ingunn/ 1
  • 002_Jarle/ 2
  • 003_Emil/ 3
  • 004_Eirik/ 4
• When read by the user recognition system the folders will be given indexes as shown in bold.
• The MMM-Facial-Recognition-2 module's users property will then be set to the following list: ["stranger", "ingunn", "jarle", "emil", "eirik"]

Running the System on a Remote Architecture

1. Run MagicMirror² on the RPI
   • Make sure MMM-Facial-Recognition-Custom MagicMirror² module is installed
2. Host the camera as a IP camera using RPi-Cam-Web-Interface software
   • Set rpi_IP in conf.json to the RPI's IP adress
   • Set run_on_rpi in conf.json to false
3. Start the recognition by running python user_recognition_main.py