Using machine learning to create a game controlled by hand gestures!
Currently, I am using a publicly available, pre-trained model called MediaPipe Hands as the image encoding backbone (extracting landmarks from hands) and training a support vector classifier using the extracted landmarks. This setup is not only accurate but also quite efficient (since MediaPipe does a good job optimizing their model!) The current "game" is barely a game and only used for testing purposes.
- Provide easy way to download dataset
- Revamp dataset collection script
- Collect new dataset with three actions
- Improve game graphics
- Frame-independent movement
- Revamp movement (simpler, smoother = better)