Playing Card Detection Using OpenCV-Python on the Raspberry Pi 3 + PiCamera

I made a playing card detector program that uses OpenCV-Python to detect and identify playing cards in a video feed. It runs on the Raspberry Pi 3 with an attached PiCamera. This video explains the image processing algorithm I use to detect and identify the cards.
I'm looking for part-time consulting or short-term contracting work in the area of computer vision! If you'd like my help implementing image processing algorithms, please get in touch with me using the email listed on my channel's About page. / edjeelectronics
Get a Raspberry Pi: amzn.to/2Iki3fb
Get a PiCamera: amzn.to/2rKxarh
The source code for the card detector program can be found at:
github.com/Edj...
More information on my Blackjack robot project can be found at:
hackaday.io/pr...
Twitter: @EdjeElectronics / edjeelectronics
----------------------
Music credit:
----------------------
Windom Earle - Kirblooey
freemusicarchiv...
Chris Zabriskie - Air Hockey Saloon
/ air-hockey-saloon
Mason Donovan - Epiphany
/ epiphany

[@EdjeElectronics]

For those interested, the card detector program now works with USB webcams. Please see the github readme for instructions on how to use it with a webcam! github.com/EdjeElectronics/OpenCV-Playing-Card-Detector

[@simplelabs7273]

hey saw your post on hackaday and the recent update... just a quick thought on the training images... how about training images with all possible pairs of numbers (13 x 13 = 169) and pairs of suits (4 x 4 = 16) . This might help with better detection of the hand when cards are dealt one above the other?

[@gonzalogoded2089]

Thank you, I am thinking on developing a program to track plays on the card-game of bridge, however the cards there are usually hand-held and often never released. Do you think it is possible to train a card detector to detect cards with fingers on the screen?

[@oxigenoplenitud6507]

do you have this for casino online ? something new bro ?

[@CharlesHollins1]

The simple and clear explanation is almost as impressive as the program. Great work!

[@stuff662]

Super well made video and crazy interesting project. I like how you explained what was going on behind the scenes in an easy to understand way!

[@DaveJones-Waveform]

Brilliant use of the camera, and impressively quick too!

[@chrisgavin]

Amazing work and video thanks. I'm just beginning with Pi, picamera and Python, but I can appreciate that this is very smart work indeed!

[@user-fg4wk3cq2d]

Many thanks for this ! Really well documented code - I'm a Pi newbie and had it working within a couple of days. I'm hoping to adapt it to make a card sorter for our bridge club..... 🙂

[@EdjeElectronics]

I'm glad you were able to get it working! You're very welcome. The card sorter project sounds cool, good luck with it!

[@user-fg4wk3cq2d]

@@EdjeElectronics I now have the code reading in a bridge notation file containing (x) deals and flagging each card identified with N/S/E/W for a pre determined deal. Next problem, I need a way of setting the hand number to be interrogated 'on the fly' whilst the CardDetector script is running. (e.g. '01' through '24' ) - any tips please ? . Thanks

[@BrianFaure1]

Great idea! Thanks for providing the code!

[@MarkJay]

Impressive that the pi can process frames that fast

[@rossmauck8254]

yeah mine was not that good, maybe he has a better picamera version

[@Kirix]

You could bring it in a casino with a hidden camera and have another electrode or vibration thing pulse when the count is high

[@slenderman4994]

THIS IS EXACTLY WHAT I NEED, the difference beeing I need to scan cards from the screen and I want to make a C++ program for Windows to do this but as I understood OpenCV is available for Windows and C++ too but I'm not that good at programming, libraries and complicated things like that...just some algorithms in console programs. -_- Do you have some tips for me, please? Sorry for my bad english :))

[@vaddix9980]

Not one dislike. Awesome!

[@davidcox8942]

You. Spoke. Too. Soon...

[@jordanknight9424]

Great demo! Have you tried classifying cards using just the top-left "icon" of each card instead of the entire playing card? This bit should have all of the information needed to classify a card and is a smaller image to work with than the full card. If the algorithm is uncertain you could then attempt to classify using the full card. Would this strategy provide a performance benefit to classification speed when running on something relatively slow like RPi3?

[@EdjeElectronics]

Thanks! The algorithm does just use the top-left corner of the card for classification. It only looks at the rank and suit in the snipped-out corner. However, it does need to use the contour of the whole card to know where the corner is.

[@Boudico]

@@EdjeElectronics This is so nice! I am wanting to learn raspi programming for a project very similar to this. Essentially, after playing a hand of "Hand Foot and Toe [canasta type]" I'd like to put the players cards in a shuffler (throttled slow) with a camera in the shuffler base to read each card corner and count the points in their hand. (especially helpful for players that are painfully slow in counting the points or perhaps are dyslexic). the next step would be assigning the proper point value to each card. As the suit is irrelevant, only the number shape is necessary. Would very much appreciate any input or point in the right direction you might be willing to give!

[@EdjeElectronics]

@@Boudico Sounds like a fun project! I haven't played Hand Foot and Toe before so I'm not familiar with the scoring system, but you should be able to do it. You can add a set of if statements that say "if card is an ace, add 15 points to the hand, if card is a king, add 10 points to the hand", and so on. Line 88 of the CardDetector.py script ( github.com/EdjeElectronics/OpenCV-Playing-Card-Detector/blob/master/CardDetector.py ) determines the rank and suit of the card in the screen. You can add the if statements after line 88. I will be posting a card-counting video that will show how to assign points when certain cards are detected. Good luck!

[@Boudico]

@@EdjeElectronics Wonderful! I look forward to it. thank you very much!

[@Boudico]

@@EdjeElectronics Progress report: about 2/3rds through the first 'python for beginners tutorial" video I found here on YT. Also got the rpi3 loaded with python 3, opencv, etc. Kept getting the "2 arguments expected 3" bit when trying to run the program and I see below (Eldrog) mentioned removing the dummy values, which I did (copied the entries and commented out the original). Got your program running as provided (thank you!). I see that it really wants a certain distance from the cards and they can't overlap. I can work with that by having the player put their cards in a 'cradle' that just fits the card so each card is placed 'perfectly'. Next step is to impatiently wait for your promised 'card counting video" :) Thanks again

[@LeveragedFinance]

Do casinos already do this with their own cameras? So when they ask for a skills check the computer can tell if you are counting?

[@EdjeElectronics]

Yes, I bet most casinos have software to do this already, especially the big casinos in Vegas. Here's a video showing card detection software running on an above-the-table casino camera... it works really well! I'm hoping to get mine to work this well eventually. kzbin.info/www/bejne/iJjNgZaGaa2Ba5o

[@parthkabariya6505]

thanks for video, I am working on same project but only deference is that i am using Go-Lang, your tutorial helped me a lot :) one more thing, video link which you have posted here, are they using Open-CV ? or tensor-flow something like that cause if cards are overlapping then it's hard to detect with Open-CV or do you know any way so we can detect card, even though they are overlapped in Open-CV

[@DarkSpeedKiller]

Is there a version in which this recognizes cards from my screen instead of a webcam or piCamera

[@EdjeElectronics]

I haven't written one that works on the screen/desktop, but you gave me another video idea: "How to Use OpenCV to Capture Your Desktop or Screen"! Actually it looks like this video shows how to do it: kzbin.info/www/bejne/fYjHo39umdlkrrM

[@DarkSpeedKiller]

@@EdjeElectronics thats nice but how can i combine it with the card detector

[@JairoMartinezVargas]

Thanks for sharing! Have you used it with domino tiles?

[@EdjeElectronics]

Nope, but it would be cool to make the program work for dominos! It could just count the number of dots to identify the tiles.

[@bigdatax6512]

please create it...

[@edderube1247]

Hi, great video and application! I am just wondering if the camera angle has a big effect on the recognition of the objects? For example the camera looking straight down vs at a 45 degree angle? Do you need to create a different set of training images at 45 degrees for yolo to work?

[@Makelifehappennow]

Great project. Well done. I am curious how you make sure the 9 and 6 are not falsely identified. Since they have roughly the same amount of pixels? Is there some other factor in the algorithm?

[@ameerhussain2051]

hey genius thumbs up for Q. like this :p

[@lrahwill9909]

That would be ceating

[@smecmanca]

Good job

[@rickk4869]

Hii, i want to run this on a print screen instead of a camera, is thar also possible by editing the code a bit?

[@erickpaulmarroquinvasconez547]

Hello, sorry, I have a question as I can edit your code so that it allows me to send a default image located on my computer ..? and no longer use streaming video ..? but what I want is to send to process a previously captured image and that throw me the results of the prediction

[@EdjeElectronics]

Hi Erick, this webpage shows you how to load images in OpenCV and work with them. I think it will help you! docs.opencv.org/3.0-beta/doc/py_tutorials/py_gui/py_image_display/py_image_display.html

[@Jsters]

this is a great program! was just wondering how the skewed 8 of spades is cropped and reoriented-I see you detect corners, how exactly is that done?

[@Pmarmagne]

It's funny because the reason I came to check out this video was to see if you could create a bot to help you with black jack

[@SinhtUpthereal]

hey i realy would pick up ur project to fullfill my project about Trading Card Games do u hafe a full documentation about ur project ?

[@F_in_the_chat]

since this is 5years old have you found any updated way to do this? having a hard time with distance of the detection and adding new images.

[@EdjeElectronics]

Yes, use machine learning instead :) kzbin.info/www/bejne/hJeWq3WAZ8h3d9U

[@nrg007]

Fantastic job done there. I was on the same lines and building a robot for poker. I am using a PS3 eye usb cam. Your code works perfectly well with black cards, however with red cards it recognizes the suit but not the rank like "unknown of hearts" or "unknown of diamonds". i do keep it on a dark background but still. I faced the same issue when i use your rank_suit_isolator. What do you think could be the issue?

[@keyurparalkar9535]

AWESOME!!!!

[@IF25497]

Hi. It's me again. I have a new question. Since using laptop's camera is kinda unconvinience, is it possible to use my phone's camera instead( connect to my laptop via cable)? Thank for helping. Have a nice day sir!

[@EdjeElectronics]

The easiest way to solve this is to use a webcam. You can find them for pretty cheap on Amazon. Here's one I use! www.amazon.com/gp/product/B01M642ZTC

[@IF25497]

@@EdjeElectronics thank for your replying. I had already been able to use my phone's camera in this program. It's working fine, higher accuracy but lower FPS when detect cards ^^. Have a nice day sir!

[@onelgates]

Can it detect or read a Picture of a random card? Example is when I put my iphone in front of the webcam with a picture of random cards, will the script recognize it? Just curious.

[@EdjeElectronics]

Hmm, it probably wouldn't work in that scenario. If the background (whatever you set the iPhone on) isn't one solid color, this program might not be able to isolate the card. Also, the lighting of the iPhone might cause the program not to work.

[@computinghub9550]

Is there a way to make the code work when detecting the computer screen, instead of a camera?

[@student747]

yes. just make screenshots...

[@gambler301]

Or obs fake camera?

[@user-fu6fs3zw3w]

hi! Thank you for your project. I ran CardDetector but got an error. I added import os to Card.py but it keeps getting errors. "error: (-209: Sizes of input arguments do not match) The operation is neither 'array op array' (where arrays have the same size and the same number of channels), nor 'array op scalar', nor 'scalar op array 'in function' cv :: arithm_op '" I do not know what should I do. Please help me. What should I do?

[@MrThomas5701]

Just delete the dummies variable in card detector py

[@IF25497]

Hi. Thank you for your study. So I have a homework that has to detect something with openCV. And I'm interested in your study. I've followed the instructions that you made to detect cards. After messing around with errors. I successfully detected cards. But there's a problem for me: Not all the time I show the card it can be detected. In your video, you need a dark background. But I'm using my laptop's cam so I just show the cards in front of it. When the lights on my room's off, the program's working just fine. But when I turn on the lights (maybe my background's lighter?), it cannot detect contours and identify cards. What should I do to make it work all the time? Thank you for reading. Hope you can help me out!

[@EdjeElectronics]

Hello, thanks for your interest in this project! This algorithm only works well against a uniform background. If there are other objects in the image, or if your fingers are obscuring the cards, it won't work as well. That's why I switched to using machine learning! You can use a machine learning object detection model to detect playing cards in more realistic conditions. Check out my other video, which shows how to train your own playing card detector: kzbin.info/www/bejne/iJjTl55smr6rgKM . It's a lot of work, though! It takes a least a few days to get it all figured out.

[@IF25497]

@@EdjeElectronics Oh It's also you. I watched that video the other day but didn't recognize you xD mybad. I was thinking of following that instruction too, but then I realized my laptop's requirements ain't allow me to so all I could do was dropping a like for you. Good work bro. Keep it up. And once again, thank you a lot.

[@piergiuseppedabbraccioTheBUC]

would it be possible to see the PC screen? So to check on images that run on the screen.

[@Lachsnacken3000]

Same question here

[@3CHONPRODUCTIONS]

Yes it easy

[@freecode.ai-]

Well at least we know it can be done now. Maybe explain how next time. Like using pillow and image grab with bounding box..

[@3CHONPRODUCTIONS]

@@freecode.ai- DM me if u want to learn

[@vidsava]

@@3CHONPRODUCTIONSi'm interesting

[@mnz1151]

Great info! I want to use with a regular webcam detect cards played in a game on a large tv from 10ft away... pi(cam)-----10ft---tv(playing a blackjack game). My goal is great stats of cards and plays...Like will most of time that I get 2 cards adding 12...would a hit will be 22? and many curios stats that I have. The game is very nice and good one that I play on prime tv app. It says has a certification for random...but I disagree...almost all double cards, split in more of same card...like two 6, split in 66 and 67 and may get a 4th 6...that situation happens many time for different ranks. But the game is really good and realistic.

[@JCB438]

Hi! I had some issues with the code, because I have a different version of CV, like cv2.findContours now gives 2 arguments instead of 3, but that's solve just by deleting the "dummy". But i can't fix an issue with cv2.absdiff(), it says "error: (-209:Sizes of input arguments do not match) The operation is neither 'array op array' (where arrays have the same size and the same number of channels), nor 'array op scalar', nor 'scalar op array' in function 'cv::arithm_op'". Could you help me please? Thanks!. Edit: I just fix it by adding "import os" to cards.py. Thanks for the proyect!

[@EdjeElectronics]

Hi @Eldrog , thanks for updating with your fix! I'm glad you were able to figure it out. Sounds like you know more about the OpenCV backend than I do anyway! :)

[@allanreylayese5349]

hi sir thanks for the tutorial i would like to ask how to achieve high frame rate we are doing our project to detect plastic bottle but the detection is slow delay about 7 seconds

[@siddhantsangal5298]

Can this be implemented using tensorflow? What changes will I have to make to the code in your video showing how to install tensorflow on the pi

[@rafaelalfonsosuarezherazo8836]

What a good job, friend. I'm doing something like this and I need some advice as I can communicate with you. please I would thank you in advance.

[@412kev2]

Yo is it cool if I edit your code to make a version that works with webcams

[@EdjeElectronics]

Yes it is! Here's a good resource showing how to access your webcam with OpenCV: pythonprogramming.net/loading-video-python-opencv-tutorial/ I don't have a webcam, so let me know if you get it to work!

[@412kev2]

Alright thanks Ill let you know

[@viduragamage9094]

412 Kev could you please share what you've done up to now with the web cam?

[@EdjeElectronics]

Your comment inspired me to finally modify the code to support USB webcams! Please see the github page for how to use it with a USB webcam. github.com/EdjeElectronics/OpenCV-Playing-Card-Detector

[@parthkabariya6505]

it is showing every time " unknown of unknown" can you help me and it only working for video which run at 10 to 12 fps

[@minijginger1718]

Is it ok to edit your code as a reference for my final project :)thanks

[@EdjeElectronics]

Yes, you are welcome to use and edit my code, thank you for asking!

[@zikrimohammad3720]

can i do this program by using only the lappy webcam?

[@salehabdullah8453]

Hi I run the code on raspberry pi using visual studio code, the code run without error and opened a live stream window but, when I put the card in front of the camera it didn’t detect the card, can you please help me in that?

[@Ihavetoreturnsomevideotapes]

You should have a patreon

[@EdjeElectronics]

Hmm... maybe someday! :)

[@user-cb6ee1ts2g]

What Cards' brand are you use?

[@EdjeElectronics]

I'm not sure what the brand is, I got them for free in the mail from State Farm insurance. You can use any brand of cards by using the Rank_Suit_Isolator.py script located at github.com/EdjeElectronics/OpenCV-Playing-Card-Detector/blob/master/Rank_Suit_Isolator.py . The script takes pictures of individual cards, and isolates the ranks and suits. If you run the script, it first prompts you to take a picture of an Ace, then a Two, and so on. Then it will prompt for a picture of Spades, Diamonds, etc. Each time you take a picture, it will save the isolated rank or suit to the hard drive. To use the card detector with your own brand of cards, just run through the script taking pictures of the cards.

[@user-cb6ee1ts2g]

Ok,I'll try it.Thanks a lot!!

[@freecode.ai-]

Could you show me how to do this in c++?

[@EdjeElectronics]

Sorry but no, it would take a very long time to convert this program into C++!

[@freecode.ai-]

@@EdjeElectronics I already did it thanks

[@EdjeElectronics]

@@freecode.ai- Nice! You must be pretty familiar with C++. I'm still a bit of a rookie with it! I haven't used the OpenCV C++ library at all.

[@freecode.ai-]

@@EdjeElectronics I am guessing it has all the same functions.. I did include all the contribs when I was building my install so it may be a little different.

[@pssa3215]

@Jessie Sasser do you still have the code ? I really need it for my course :)

[@marcinkurzydym288]

Hi. How can I open the windows wich is called Transform and a Rank window even the Rank Sized. And ofc if it's possible how to get the window with Matching query and Match quality in 3:50. Thank You! :)

[@onelgates]

Can I run this in orange pi?

[@SamratDuttabdn]

Can I run it on Raspberry pi 2?

[@EdjeElectronics]

Yes it should work on a Raspberry Pi 2 as well!

[@onelgates]

Hello. Can I run it on orange pi?

[@jogithajohn5689]

thanks for the video, but how to identify and name the object? how to train for that

[@EdjeElectronics]

Hello, check out my other video, which shows how to train a neural network to detect custom objects! kzbin.info/www/bejne/iJjTl55smr6rgKM

[@micomator]

2:22 that's not the eight of spades yo

[@ilhamakbar5791]

how to detect the distance from camera to card ?

[@awaisahmadsiddiqi6505]

i think you need to calibrate it first. for example, taking a card 2 inches away will shorten the size by how much percentage. by doing this, you can approximate. however for this example, it doesn't actually matter because we are finding contours first. and the inner text will be cropped out proportionally

[@thebulat4139]

Hi. I have issue. When a card enter the camera's sight it displayed this error: OpenCV Error: Sizes of input arguments do not match (The operation is neither 'array op array' (where arrays have the same size and the same number of channels), nor 'array op scalar', nor 'scalar op array') in cv::arithm_op, file C:\projects\opencv-python\opencv\modules\core\src\arithm.cpp, line 659 Traceback (most recent call last): File "C:/Users/Haziq/PycharmProjects/CardDetector/CardDetector.py", line 86, in k].suit_diff = Cards.match_card(cards[k], train_ranks, train_suits) File "D:\OpenCV-Playing-Card-Detector-master\Cards.py", line 261, in match_card diff_img = cv2.absdiff(qCard.rank_img, Trank.img) cv2.error: C:\projects\opencv-python\opencv\modules\core\src\arithm.cpp:659: error: (-209) The operation is neither 'array op array' (where arrays have the same size and the same number of channels), nor 'array op scalar', nor 'scalar op array' in function cv::arithm_op I also posted this issue in the github. Thanks in advance.

[@maartendebaecke7228]

Remove the "dummy, " on line 261 and do the same thing for all the other lines that start with dummy.

[@manhdoan199x3]

how to run code ?

[@LeveragedFinance]

Don't bring that into an actual casino. For your safety