Installation process for facial recognition dependancies

fr_deps.md

Step 1: Install dependancies

sudo apt -y update && sudo apt -y full-upgrade
sudo apt install build-essential \
    cmake \
    gfortran \
    git \
    wget \
    curl \
    graphicsmagick \
    libgraphicsmagick1-dev \
    libatlas-base-dev \
    libavcodec-dev \
    libavformat-dev \
    libboost-all-dev \
    libgtk2.0-dev \
    libjpeg-dev \
    liblapack-dev \
    libswscale-dev \
    pkg-config \
    python3-dev \
    python3-numpy \
    python3-pip \
    zip \
    python3-picamera

Step 2: Updates

sudo pip3 install --upgrade picamera[array]

Step 3: Increase the swap file size so we can build dlib

sudo nano /etc/dphys-swapfile

Find CONF_SWAPSIZE and change its value from 100 to 1024. Save and exit then run this command:

sudo /etc/init.d/dphys-swapfile restart

Step 4: Build and install dlib

cd
git clone -b 'v19.6' --single-branch https://github.com/davisking/dlib.git
cd ./dlib
sudo python3 setup.py install --compiler-flags "-mfpu=neon"

This may take a significant time to run (Raspberry Pi 4 took about 30 minutes)

Step 5: Revert the swap size

sudo nano /etc/dphys-swapfile

Find CONF_SWAPSIZE and change its value from 1024 to 100. Save and exit then run this command:

sudo /etc/init.d/dphys-swapfile restart

Step 6: Install face_recognition and examples

sudo pip3 install face_recognition

You're done!

Hello and thank you for your work!
I'm trying to install on a Pi 4 running 32-bit Bookworm (I was originally trying with the 64-bit version, but switched to 32 when I started having problems because I understand software compatibility to be greater in the 32-bit version).

As per the "Python on Raspberry Pi" section of this support doc on the raspberry.com website, you can no longer use pip to install python packages systemwide and you must use a virtual environment. This also prevents you from being able to use sudo when doing pip install.

It also seems to be the case that in step 1, python3-picamera does not exist in the apt library, but python3-pycamera2 does. When I add the 2 to the package name, drop the -b 'v19.6' from step 4, and do everything inside a virtual environment (created using venv) and not using sudo, I get all the way through to Successfully installed face-recognition-models-0.3.0 face_recognition-1.3.0

An important amendment to my comment above which I neglected to include: Be sure to create your virtual environment with the --system-site-packages flag or else you won't have access to e.g. libcamera inside the venv.

I just installed this on a Pi zero 2 W with a Lite install of Bookworm on it. Basically the same as my previous comment except:

1. I think that if you are using picamera2 you don't need to do step 2. There is not an extra feature in picamera2 called array so the install doesn't work if you try to do pip3 install --upgrade picamera2[array] anyway. I'm not sure here, but I had trouble with the install and just skipped this step and everything else worked.

2. Step six, sudo pip3 install face_recognition appears to want more memory than the Pi zero has (it just comes with 0.5GB) so do step 5, reverting the swap size) after step six instead of before. It takes a while but eventually the installation goes all the way through.

Impressively, the Pi zero 2 can do at least basic face detection. I got the examples from git clone --single-branch https://github.com/ageitgey/face_recognition.git and updated face_recognition/examples python3 facerec_on_raspberry_pi.py to use picamera2 and cut out the recognition part so it just looks for generic faces and it can do it! I had the image size set to 640 x 480 and it processes about one image every five seconds. When I dropped that down to 320 x 240 it peps up and can do about one image per second.

Thanks again!

Hi. Amazing that a Zero 2 can do this. Thanks for your great work and for taking the time to update this. Sadly, if I spent all my time keeping these tutorials up to date with OS releases, I'd never be able to do anything else!