Handwrite the Korean alphabet, Hangul, like a native.
BTS Magic Wand lets you improve your Hangul handwriting! Using an Arduino, IMU sensor, and TensorFlow Lite for Microcontrollers, we trained a tiny machine learning model to recognize the Korean alphabet, Hangul, you are writing.
For the handwritten Hangul dataset, we wrote consonants and vowels of Hangul in various sizes (e.g., small, medium, and large) with an Arduino-connected pen. Using the Arduino board, we collected three-axis gyroscope data per alphabet. Then, we trained a CNN model using the gyroscope data.
If you start writing Hangul with the Arduino-connected pen, this experiment first captures IMU data from the board. Then, based on a pre-trained CNN model, it labels the captured data with matching Hangul alphabet. Through the text displayed on the screen, you can check and correct your Hangul handwriting.
Ultimately, with BTS Magic Wand, you can write and learn Hangul more accurately with your own hands.
Try BTS Magic Wand to write BTS members' names in Korean! 🇰🇷
- Microcontroller: Arduino Nano 33 BLE (with IMU sensor)
- Tensorflow 2.0 (tf.Keras), TensorFlow Lite
- CNN / LSTM, Network Optimization, Post-training Quantization
- Your own deployment platform (Arduino, Sparkfun Edge etc.)
- Arduino IDE installation
tensorflow>=2.4.0
,
-
Install libraries on Arduino IDE.
- Essential library for your own deployment platform (In our case,
Arduino LSM9DS1
by Arduino) Arduino_TensorflowLite
by Peter Warden
- Essential library for your own deployment platform (In our case,
-
Clone the project.
# Compiled files for Arduino exists in this directory. cd blob/main/src/magic_wand/arduino/examples/magic_wand open main.ino
-
Upload the project on your device and run it !
If you have any issues, Plz report us via this channel.
https://bts-micronet.github.io/
BTS-magic-wand
is licensed under the terms of the MIT License.
Copyright 2021 BTS-MicroNet. https://bts-micronet.github.io/. All Rights Reserved.
Big Congrats. to BTS 'Butter' and 'Permission to Dance' have won Billboard 1st 🏆
- tensorflow github
- tflite-micro github (Similar with original tensorflow github, some renewal updates for
micro
parts for tensorflow lite are included.)