A python 3 application that will lock the computer screen when you leave, and the camera won't detect your face in front of it anymore for a specified interval of time.
Locking the computer automatically is always painful. You can put a 10 second screensaver timeout, but if you stop using the mouse and the keyboard, for whatever reason, it becomes pretty annoying pretty fast. In the past, to protect my sessions from intrusions I wrote a script that locked the screen when the signal strength of my bluetooth devices (phone or smartwatch) would drop under a specified threshold (typically, when I walked away). This worked well as long as I had a Linux workstation, now that I have been forced on a macbook I really do not want to learn ObSolete/X apis in Objective C.. I can come with better ways to waste my time.
Thinking and rethinking about zero-hassle ways to do this, I had a new idea: have the camera detect constantly my face in front of the screen, and when it does not see it anymore, lock it.
This is based on dlib, a machine learning C++ library that has incredible facial recognition features. In Python you can find a package called face_recognition that wraps dlib. Python was a natural choice to avoid having to do this in C++, as I am not really keen on the idea to learn any OSX specific stuff. The software goes in background and opens a Qt5 gui using a system tray icon. It uses opencv to read the camera frames, and it will perform a face recognition on the camera frames against a reference image. When the target face is not present anymore in the camera frames, it will launch a configurable command (typically to lock the screen).
One of the main selling points of dlib is that its face recognition algorithms do not need training: just one image will suffice.
Warning: On the first run facelock will check that there is a proper face image set in the preferences, otherwise it will prompt you to select one. It will similarly warn you of a missing external command to lock the screen in the settings (see below for how to set this on specific OSes). The image should be a clear front view picture of you, and that is usually enough. You can tune it and try different pictures (set them on the Settings panel) and the Calibration utility.
Included in the repository there is a small utility called 'Calibration'. This can be launched from Facelock main menu, and will show a visual feedback on the face recognition process: it can be used to verify that the camera is working properly, the image is not doo dark, or the best angles to position the laptop or the camera.
To exit from the calibration window, press the 'q' when the window is in focus.
The prerequisite is to have the brew package manager installed.
(NEW) Consider the install.sh script in the osx folder, that will try
to do all the necessary installation steps automatically. In case of problems,
refer to the description of the necessary steps in this paragraph.
To use the script, go to the facelock git repository that you have cloned,
then type
cd osx
./install.sh
Manual installation
First thing you need python 3:
brew install python3
Then install opencv
brew install opencv
Opencv is best installed from the package manager brew. If you don't and
use pip install opencv_python instead, this version will bring a private
installation of the qt5 libraries which will conflict with the ones installed
regularly.
Finally, install the proper requirements from the facelock directory
sudo pip3 install -r requirements_osx.txt
and this should be about it.
Run it with
python3 ./facelock.py
In the osx/ folder there is a utility called makeapp. Using it, it's
possible to generate a proper OSX app package, to launch the
application from the finder or pin it to the dock. Everytime that the
app is updated from git makeapp must be used to regenerate it.
cd osx
./makeapp
There is an experimental script that will attempt to disable the python rocket icon by detecting the Info.plist descriptor of the python launcher app and inserting an LSUIElement set to true.
In order to start it,
cd osx
./undock_python.sh
The original Info.plist file will be backed up in the current directory in Info.plist.bck .
WARNING:
when you press "stop tracking" from the facelock menu,the python script will try to free the resources allocated from opencv with VideoCapture.release(), as per opencv's examples. While this works perfectly on linux, on OSX it will often crash the application with a SIGSEGV or a'NSInvalidArgumentException', reason: '-[NSTaggedPointerString unlock]: unrecognized selector sent to instance
This seems to be a problem of opencv.
The linux installation should be pretty straightforward. Use your package manager to install python3, for example with
# dnf install python3
(fedora)
# yum install python3
(redhat)
# apt-get install python3
(debian/ubuntu)
after this, running
# pip3 install -r requirements.txt
should do the trick. Your mileage may vary according to the distribution used, especially for PyQt5 and opencv-python: in some distributions it might be preferable to install those dependencies through the package manager and not pip.
NOTE: on some desktop environments, like Gnome 3 Shell , you might need some extra extension to be able to see the facelock icon.
For example, on fedora with Gnome 3, run
sudo dnf install gnome-shell-extension-topicons-plus.noarch
On Enlightment, select "Create new Systray" from the desktop menu.
The first time that facelock runs, it will generate a configuration file in a OS-compliant conventions fashion. Since the imagepath configuration will be empty, you will be prompted to select an image.
At first run or when the 'command to execute setting' is empty, facelock will automatically configure a path to the lockscreen.sh script, that will lock the screen in a multiplatform fashion (see paragraph).
Have a look at the settings panel and start the tracking from the facelock popup menu.
Delay time : The timeout to lock the screen if the target face is not seen for the the specified amount of seconds.
Command to execute : A command to execute. Can be any command, it defaults to <FACELOCK_HOME>/lockscreen.sh .
Immediately begin tracking at application startup : This will start looking for the target face as soon as the application starts.
Immediately lock if tracking only unknown faces : This allows you to have a higher timeout, but lock instantly if an unknown face is detected in front of the computer.
Warning: this could have security consequences, especially on high timeouts.
Image Path : The target face image.
Target face name : a label for the target face. It will be shown when
doing calibration, in an highlighted red box surrounding the target face.
Non matching faces will be labeled as "unknown".
Processed frames # : This will start the face recognition every n frames out of the maximum camera frame rate per second. It is useful to impact less on the cpu.
This script is the default command to be launched on a timeout. It will try to detect the enviroment in which it is running and launch the according command to lock the screen.
So far, the working environments are
- Gnome
- Kde
- Mate
- Cinnamon
- Aqua (macOS' desktop)
In all the other cases it will try to fallback on the generic command xlock,
that will therefore have to be available on the path.
For example, on fedora, do
sudo dnf install xlockmore
to obtain it.
Make sure that for each environment you need, the specific lock screen commands are installed, e.g. cinnamon-screensaver, mate-screensaver, etc.