Execute some commands when your face is not detected!
My work place ComplyAdvantage has this 'policy' that if you leave your laptop unlocked and someone else manage to send an email to the whole company from your account, you then need to buy donuts for the company.
I fell victim on the second day upon joining!
I was so upset that I have to do something, I knew it won't be my last time to get 'donut-ed'.
I first decided to use openCV's face detection to check for the absence of a face via my front webcam - If there is no face detected in X seconds, my screen will lock.
It works, but I am not very satisfied, since it is a face detection algorithm, any face will do. If any of my colleague quickly get in front of my laptop before it locks, then it fails.
What I need is face recognition. I certainly don't have millions of images and the GPU power to train a robust AlexNet or ResNet. So instead I spent sometime to integrate the Microsoft Cognitive Face API to my script.
- Use Python 3.6+
- Get a FREE Microsoft Azure account, and get a API key for Face API in Cognitive Service.
- Rename facelock.cfg.template to facelock.cfg, add your KEY value to the file, and add any commands you would like to execute, for detailed usage, refer to python's subprocess, for example, to logout Google services in MacOS:
COMMANDS = [["open", "-a", "Google Chrome", "http://accounts.google.com/logout"]]
- You need to 'train' a reference face (your face) from an image (JPEG etc), please use a good quality picture showing your front face. The image can be a local file or from the web:
python facelock/facelock.py train [OPTIONS] IMAGE_LOCATION
A model.pkl file will be generated.
- Run (with default paramters):
python facelock/facelock.py verify
use
python facelock/facelock.py verify --help
to get the options
If the reference face is absent for a set amount of time (default 30 seconds), the script will execute a list of commands specified in the config file.
Microsoft's Free account has a 20 API calls/minute request limit. For Face recognition, it takes two API calls, (Detect + Verify), that means the maximum sampling per minute is 6. If you use a more frequent --sample-interval value you will get 'Too Many Requests' Error.
The Face Detect API call returns a faceId, it is valid for 24 hours only, after this, you need to re-run
python facelock/facelock.py train