amfl / opencv-go

Computer vision helper for the board game Go/Baduk/Weiqi

OpenCV-Go

Point a webcam at your Go board and record real games in sgf format.

The resulting sgf file could be analyzed with AI.

Status

This project isn't really in a robust, working state yet.

It's written by me for fun, so don't expect state-of-the-art techniques or incredible documentation while it all comes together.

Usage

python3 main.py <VIDEO_FILE_OR_DEVICE_NUM>

Examples:

python3 main.py 0              # Opens video device 0
python3 main.py /foo/bar.mkv   # Opens a video file

Thoughts

Random thoughts on how to track a Go board

• Figure out roughly where the board is based on color, get convex hull
  • https://docs.opencv.org/master/d7/d1d/tutorial_hull.html
• Obtain matrix from hull which maps some abstract grid onto board
  • find homography?
  • Use extrapolated hough lines instead of convex hull? Might work better with occluded corners, or skin tone matching board color.
• Downres image pretty hard
• Sample pixel color at transformed grid point