Point a webcam at your Go board and record real games in sgf format.
The resulting sgf file could be analyzed with AI.
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.
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 fileRandom thoughts on how to track a Go board
- Figure out roughly where the board is based on color, get convex hull
- 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