The front of the final product with the camera visible
The side/rear view of the case with the (covered) LED display
A diagram outlining the classification algorithm for each card
The microcontroller board (an OpenMV H7) inside the case
The LED display mounted inside the case
Project information
- Category: Embedded System, ML
- Completion date: March 2023
- Technologies: STM32, CNN, Image Processing
Details
For my DSP capstone project at UCLA, I created an embedded system that localizes and classifies 12 cards for the game SET. I built this system with my teammate, Tyler Price, to allow someone to improve their performance in the game SET using this handheld device. The final product has a very simple user interface with a camera on one side and an LED display on the other that displays where the matching sets of cards are located.
The system has algorithms to localize each of the 12 cards in play and then classify each card's 4 attributes: type of shape, number of shapes, shape color, and shape fill. Each attribute is classified using a different algorithm to create the most efficient system possible. The algorithms utilize several image processing techniques such as filtering, CNNs, and color distribution analysis.
Our project report is available here and my project blog is here.