Made by david-0609, for the FSFE YH4F Coding Competition 2021
This program reads in coordinates of finger landmarks (see landmarks.png) and uses the data to find fingters that are up and by tracking the movements of the fingertips, detects a gesture and executes a keyboard shortcut linked to it. A premade module for hand detection is used and OpenCV is used to record dta from the webcam. The program is designed modularly and uses a facade design pattern, with the module Run being the facade object. The library being used to execure keystrokes is pyautogui, which can only be used under a GUI environment. Configparser is used to read in the config file, see https://docs.python.org/3/library/configparser.html for detailed explanation of the format. The algorithm used to detect the direcion of travel of the finger is quite simple and can be improved to use a more sophisticated method for a better result, for the purpose of this project, the accuracy is sufficient.
The project works most of the time, however, due to the impercise data from of the hand detection algorithm and lack of fine tuning, it is currently not guranteed to be always accurate in detecting hand gestures effectively. If the input is perfect, the code would able to reliably recognise the gesture and execute actions based on the detection results.
- Python 3.7, other versions of Python may not be compatible with the libraries, as of now, newer python versions (>=3.7) can be used but not tested
- X11 (Needed for pynput and pyautogui module), Wayland currently not tested
- Numpy, OpenCV, mediapipe etc (see requirements.txt)
- Works on any modern GNU/Linux distro, Windows and Mac not tested
- Webcam (Recommended minimum 720p 30fps) and good lighting conditions
- At least 200-300 MB of free RAM
pip install -r requirements.txt
It is recommended to create a virtual environment to install the python packages.
1. The model will draw a covex hull around the hand after separating the hand from the background, using the convex hull, find fingertips and track for gestures (See experiement1.py) As this approach gave highly inconsistant results, this method is scrapped.
Algorithm to use: https://google.github.io/mediapipe/solutions/hands.html
The mediapipe module will grab the coordinates of the points on the hand, and these points will be used to determine if a finger is being held up or not. Motion tracking will be done with numpy and matplotlib by logging the coordinate changes of the fingertips inside a detection window that is triggered by counting the number of fingers up. OOP will be used for extendiblity in the future and ease of access with a front end TUI/GUI application.
A detection window will be started as soon as all 5 fingers are found on screen, the default value is 3 seconds. Frames in the future 3 seconds will be monitored and after the finger is out of the camera or the window is over, an action will be performed through keyboard shortcut based on the results of monitoring. The x and y coordinates will be monitored and determined similiarly to the finger_up function.
See Repo
Running with debug without specifying config file (defaults to .config in home directory)
python run.py --debug=true
Running without debug
python run.py
Many Thanks to:
- My parents, who supported me throughout this project
- brokenbyte, who gave me lots of tips on development
- Tristan, who helped me develop my idea
- My friends, who gave me ideas for extra features
- Of course, there is always StackOverflow