https://github.com/laserborg/dobblebot

image-matching Dobble Card Symbols
https://github.com/laserborg/dobblebot

Last synced: 8 days ago
JSON representation

image-matching Dobble Card Symbols

• Host: GitHub
• URL: https://github.com/laserborg/dobblebot
• Owner: LaserBorg
• License: gpl-3.0
• Created: 2022-01-14T23:19:45.000Z (about 3 years ago)
• Default Branch: main
• Last Pushed: 2022-01-21T12:48:16.000Z (about 3 years ago)
• Last Synced: 2024-11-15T19:22:38.901Z (2 months ago)
• Language: Python
• Size: 19.9 MB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# DobbleBot

image-matching Dobble Card Symbols.

One fine day it might work on a real-time camera feed and win against people like DeepBlue and AlphaGo did. 

## Dataset

You can download all the reference images from this [kaggle competition](https://www.kaggle.com/grouby/dobble-card-images)

, or scan them on your own.

### * WORK-IN-PROGRESS *

...

#### What it does until now:

``01_circles.py`` is used to digitize the dobble cards if you decide to create your own dataset.
It utilizes *hough circles* (Tutorial [here](https://www.pyimagesearch.com/2014/07/21/detecting-circles-images-using-opencv-hough-circles/), somewhat explained [here](https://www.geeksforgeeks.org/circle-detection-using-opencv-python/)) by splitting the circles out of Scanner Images (I used DIN-A4 @ 600dpi).


For further reading, this would be an alternative to [find Circles by their shape](https://www.geeksforgeeks.org/find-circles-and-ellipses-in-an-image-using-opencv-python/) instead using Hough.

``02_symbols.py`` is used to crop the actual symbols out of the circles.


I used functions for [finding blobs](https://learnopencv.com/blob-detection-using-opencv-python-c/), 

then [finding contours](https://pythonexamples.org/python-opencv-cv2-find-contours-in-image/) 

(e.g.[depending on brightness](https://www.pyimagesearch.com/2016/10/31/detecting-multiple-bright-spots-in-an-image-with-python-and-opencv/)),

then [get the bounding box](https://docs.opencv.org/3.4.15/da/d0c/tutorial_bounding_rects_circles.html) to crop the image down.

``03_featurematching.py`` extracts SIFT features of all 64 template images (once) and stores them as pickle-files for later use.


![template feature](images/results/templateFeatures.jpg)

It then extracts features of the (live) image


![image features](images/results/imageFeatures.jpg)

and identifies symbols by matching features (and even computes the homography matrix -> perspective displayed as blue square).


![match](images/results/dobble_01_1000%20matches.jpg)

### what it does not (yet):

* keep a list of which symbols you have on your own card

* iterate over the "foreign" card and try to match all symbols that your card has, identifying the matching symbol (there can only be 1).

### what would be stupid and fantastic at the same time:

* calculate 3D translation matrix from homography and camera intrinsics for each symbol.

* calculate normal vectors from translation matrix. Coplanar symbols are on the same card.

* abuse code to make something Augmented-Reality-ish out of it.

* optimize image matching with deep learning instead of SIFT feature matching.