Noah Schuller - coding Project 1

Chess Piece Detection

First, I took over 3,000 photos of my chess pieces using my smartphone. These images were then uploaded to Roboflow, where I annotated the individual pieces by drawing bounding boxes around them. Additionally, I created augmentations, meaning the images were automatically modified slightly to increase the dataset with digitally altered variations. The images were then split into a training set and a validation set. With this data, I was able to train a YOLOv8 model via Google Colab, enabling it to recognize the chess pieces in real time.

Next, I corrected the perspective of the live feed from my camera and overlaid a digital chessboard onto the real one. This allowed me to check whether a piece was within a specific square, enabling the system to accurately assign and output the positions of all the pieces. Feel free to check out the Roboflow page and to improve my model.

© 2024 Noah Schuller