This is a Raspberry Pi project that was completed at BlueStamp Engineering in July 2021.

View the Project on GitHub tyzhou05/BlueStamp2021

Chess Board Detection with TensorFlow and Raspberry Pi


This project uses a Raspberry Pi to detect and re-format chess boards, and it was created in 3 weeks through an online program at BlueStamp Engineering. Overall, I was able to implement two different programs, which allow my Raspberry Pi to detect general images, and also convert Chessboards into more simple formats. I learned tons about Raspberry Pi, machine learning, debugging, and engineering in general!

Engineer School Area of Interest Grade
Tony Z. Valley Christian High School Computer Science/Engineering Incoming Junior

Final Milestone

After looking at different ways to create and train a custom object detection model, I soon realized that it would take too much time. I then looked online, and was able to find a very interesting model that took a chessboard, compressed it, and converted it into an online format using a neural network in TensorFlow. Over the next few days, I worked on training and testing this custom model, and once I finished, I found out that the FEN(notation system for chess boards online) algorithm was private, and I realized that trying to create my own algorithm would take too much time.


Before detection


After detection!


The model training and running, showing accuracy

Second Milestone

For my second milestone, I first started by following this article to set up TensorFlow, an open-source machine learning database for object detection. I soon ran into some problems involving downloading the now updated version of the TensorFlow model through GitHub, which had been modified extensively since the article was published. I spent multiple days debugging and trying to figure out solutions to it, and I was finally able to find a much better tutorial, which allowed me to detect objects through images and videos! I then was able to set up my Raspberry Pi Camera, and the software also worked through a live feed and was able to detect objects in my room. Although Google’s public detection model works well for common objects, there were still some errors(such as detecting a bird feeder as a fire hydrant), so for my next milestone, I am planning on creating and training my own object detection model.


A pretty good detection!


Another pretty good detection!

First Milestone

My first milestone was being able to setup my Raspberry Pi. I was able to unpack my materials, and first download Raspian to an SD card for my Raspberry Pi. I then connected my Raspberry Pi to power, a second HDMI monitor, and an Amazon Basic keyboard and mouse set. I then was able to set up VNC, a software that allows me to view my Raspberry Pi from a window on my main laptop monitor instead of through my second monitor, which makes life a lot easier! For my next steps, I plan to install TensorFlow and get started with object detection.


My Raspberry Pi setup


My second monitor setup