Rubik's Cube Face Detection

An OpenCV-based computer vision project to detect the nine coloured stickers on each face of a Rubik's cube.

demo.mp4

Installation and usage

git clone https://github.com/exactful/rubiks-cube-face-detection.git
cd rubiks-cube-face-detection
python3 -m venv venv
source venv/bin/activate
pip3 install -r requirements.txt
python3 detect.py

Strategy

• Get the current frame from the webcam

• De-noise the image and detect vertical + horizontal edges

• Detect the contours

• Iterate the list and retain the contours that meet these requirements:

  • Polygon with four sides
  • Width and height are "squarish"
  • Width is between an upper and lower threshold
  • Area is greater than some threshold
  • Dominant colour is close enough to be one of the sticker colours

If the list contains nine contours after this step, there's a pretty good chance we have found the nine coloured stickers. However, we need to check the position of the contours to ensure they are distributed within a sensible distance of the contour in the centre.

• Order the contours based on their x, y position and find the centre contour

• Check the contours that are furthest to the left and right, and furthest to the top and bottom, are within a reasonable distance of the centre contour

• If yes, great. Draw the contour outlines and the colours detected on the face

In action

Points of interest

• Uses the OpenCV findContours method to detect contours
• Uses KMeans clustering (k=1) to identify the dominant colour in a contour
• Uses the CIEDE2000 colour difference to detect and match colours
• OpenCV requires BGR not RGB because reasons

Todo

• Option to calibrate colours to improve colour matching in different light conditions
• Solve the cube after capturing six sides; not a computer vision problem so not intending to do this