Embedded Files
Project Description: Inspired by The New York Times' Wordle, I designed and built an interactive physical version of the game. A machine learning model recognizes the letter tiles placed on the board, while the system evaluates each guess according to the official Wordle rules and illuminates the corresponding tiles with colored LEDs for real-time feedback.
Demonstration of Letter Recognition and Light-Up Sequence
Wordle cropped.mp4Computer Vision/Machine Learning
Using an overhead camera and the OpenCV library, an image of each of the letters on the board is fed into a neural network. A training set of 1000+ images was obtained, resulting in a 97% accuracy.
Processing Pipeline
Training Data Collection
To account for slight variations in the overhead camera position, I created a letter localization sequence. The board illuminates each tile one at a time, allowing the system to detect each tile’s position, capture an image, and crop it for training data. The video to the right demonstrates this process.
Wordle cropped actual.mp4System Details
CAD Design
Designed the board and letter tiles in Solidworks.
Used SolidWorks macros to automatically generate 26 unique configurations, one for each letter of the alphabet, reducing repetitive CAD work.
3D printed all components.
Embedded System
Programmed an Arduino using the Arduino IDE to control an addressable LED strip.
Established serial communication between the Python application and the Arduino, allowing the predicted letter to illuminate the corresponding LED.
Python
Developed the primary application in Python using OpenCV for image acquisition and processing.
Utilized TensorFlow for machine learning and letter classification.
Page updated
Google Sites
Report abuse