phluxx's profile - activity

Your question sounds like blurryness is not really the property you are searching for. Blur means that pixels colors are smearing into their neighbours, so you never have big differences in color values for neighboring pixels. Typical blur detection takes that into account, it always measures differences between pixel neighbors.

Raindrops or smudges are not blur per se (except when they are directly on the camera lens). You might have pretty sharp raindrops, in fact. Probably you should start with a collection of typical examples which should be detected as "blurry" and categorize them. Fog is clearly blur, so a simple blur detection may be one component. Maybe you need to detect artifacts as raindrops in a different step?

Hi!

I am a quite experienced programmer with some math background, and very limited knowledge of CV (had a cource once at university, know the concept of feature), but I never worked with OpenCV.

I am thinking about a little private project. First step would be detection of a tabletop game setup of Catan, which you can see here: https://tametheboardgame.files.wordpr...

This image shows to seafaring extension, of which I don't care about, so forget about the blue stuff on the border. Also I would not care for the items representing streets (wooden bars) and settlements (wooden houses) for now.

What I would like to realize in a first and second step:

1. detect the hexagonal cards defining the playfield
2. detect the chips with numbers

Detection should work with better consumer level web cams (full hd), arbitrary light conditions and from arbitrary camera positions. However, there is no motion, so at least exposure times are not a problem.

Could someone point me into some directions to go for this? Are there beginners tutorials which will lead me directly where I want to go? Or could someone tell some core concepts I should stick with?

Implementation might start with python as a play stage, but would be done in c++ finally.

Thanks;)

Addendum: probably it would be the best to detect the basic geometry first (where are the cards positioned, which orientation.) Thinking about this would probably change my plan to:

1. detect a group of hexagonal tiles in the image.
2. classify each tile into one of the possible card types
3. detect the number chips (later)