Drawings matching has been developed leveragning two naive approaches:

• template matching with CCOEFF
• keypoint detection and matching with SIFT

1. Place all the objects you want to match to the corresponding folder in data/object
2. Specify the path to the target image in the `config.yml'
3. Prepare and activate dedicated virtual environment via venv or conda
4. Install requirements via python3 install -r requirements.txt
5. Run matching_ccoeff.py to obtain matching results leveraging CCOEFF
6. Run matching_sift.py to obtain matching results leveraging SIFT
7. Check output/ folder for the csv file with detections and png image with drawings
8. Play with parameters in the config.yml and choose the best parameters for your task

The implementation is fairly straightforward.

After some trial and error, it was possible to get some meaningful matches via SIFT.

The one thing that boosted it the most was the angle_tolerance parameter. I introduced it in order to remove shapes that do not look like a rectangle (since our objects are very close to a rectangular shape). This helped to remove all false positives. The rest was pretty trivial for matching tasks.

The rest of the stuff was pretty trivial for matching tasks and probably doesn't need to be mentioned.