About

Face recognition on live camera images. Runs smoothly on NVIDIA Jetson Nano (camera image 25 frames/second, face recognition updated 2 times/second).

What can you use this for?

I use it for the Face recognition game: Letting people explore what parts of a face are important for a face recognition system by making them wear funny masks (see gif).

You can also adapt the code and use it for all other kinds of face recognition / live computer vision demos.

Background

This project was inspired by Adam Geitgey’s fantastic tutorial Build a Hardware-based Face Recognition System for $150 with the Nvidia Jetson Nano and Python.

I have written a blogpost about my motivation for building the Face recognition game.

Compared to Adam Geitgey’s implementation, I added/changed a bunch of things (Reasons for these changes):

• multi-process implementation to make everything run more smoothly
• button press to add person to face database (rather than automatically adding them first time they look into the camera (which does not work all that well))
• stars to visualise match quality

Pre-requisites

You can run this on a computer (with built-in camera or USB camera) and on the NVIDIA Jetson Nano with Rasperry Pi camera (hardware description).

Installation (on Jetson Nano)

TODO, for now, follow Adam Geitgey's installation instructions.

sh download_models.sh

Installation (on normal computer)

Python >= 3.7

sudo apt-get install libopenblas-dev liblapack-dev 
pip install -r requirements.txt  
sh download_models.sh

Configuration

cp config.yaml.sample config.yaml

I want to use an internal or USB camera

In config.yaml, set source: camera and change the source_setting -> camera->location entry to the id of your camera. You can find the camera id by trial and error, try -1, 0, 1, etc till it works,

I want to use the Rasperry Pi camera on my jetson

In config.yaml, set source: camera_jetson.

I want to use a pre-recorded video

In config.yaml, set source: prerecorded and change the source_setting -> prerecorded->location entry to the file path of the video.

I want to configure all kinds of other things

There is lot's of documentation directly in config.yaml.

Startup

python main.py

Keyboard commands

q: quit
c: reload the config from file (if you changed it)
d: toggle debug mode
r: register person(s) 
u: unregister last registered person(s)

You can change these in game/game.py.