Handwritten Digit Recognition Virtual Pen

I had a crazy idea of creating a “video game” using the Raspberry Pi 4 and a cool machine learning algorithm working on the backend. Happens that the idea was too big to execute fast and I got a bit frustrated trying to make everything perfect. So I decided to just divide it in small tasks and keep you updated, perhaps you have better ideas on how to optimize this.

Basically I want to be able to complete and verify a sudoku by virtually writing with a pen. The mechanism is a bit more complicated than the description.

I managed to build the first part, I adapted the Virtual Pen and Eraser created by the amazing team behind LearnOpenCV and created a CNN model to recognize the digits. Basically I write with my virtual pen a number and I can make my little processor predict which number I wrote. Sounds cool huh?

Thing is, it was a bit complicated to match matrices and dimensions ~ but it works okeish! Anyways, I want you to take a look at it, play around with it and send me some great input so I can make this idea real and fast.

And yes I did the video on my Mac cause I couldn’t manage to screen record my raspberry pi while executing cv2 🙄 but I built it and test it on the Pi and works crazy cooooool 🤯

What do you think?

GitHub: https://github.com/coding-ai/raspberrypi_handwritten_recognition

▶️ 3Speak