Playing Card Detector

Real-time playing card detection using YOLOv8, OpenCV and treys. Detects card rank and suit, counts cards on screen, and displays live bounding boxes via webcam or video feed. Includes a poker hand evaluator that identifies your best hand and estimates win probability against a random opponent.

Dataset

• Dataset used for training from Kaggle

Detect Example

 

Poker Notebook Example

 

Features

Live Detection (detect.py)

• Detects all 52 playing cards (rank + suit) in real-time
• Bounding boxes with confidence scores
• Full card contour detection using OpenCV adaptive thresholding
• Temporal smoothing across frames to reduce flickering
• Real-time card counting

Poker Hand Evaluator (PokerAnalyzer.ipynb)

• Place community cards in the top half of the frame, your hand in the bottom half
• Automatic zone detection splits cards by position using a visual midline
• Evaluates best 5-card poker hand (High Card → Royal Flush)
• Estimates win probability via 1000-hand Monte Carlo simulation against a random opponent

Requirements

• Python 3.8+
• Ultralytics YOLOv8
• OpenCV
• NumPy
• Matplotlib
• treys

Installation

git clone https://github.com/Sxres/PlayingCardsDetection
cd PlayingCardsDetection
uv sync 

Usage

Webcam:

python detect.py

Poker Hand Evaluator:

Open CardDetector.ipynb and run through notebook:

In your image, place your 2 player cards in the bottom half of the frame and 5 community cards in the top half for the dealer.

Project Structure

PlayingCardsDetection/
├── detect.py
├── PokerAnalyzer.ipynb
├── YoloCardTraining.ipynb
├── Runs/
├── models/
│   └── cards.pt
├── uv.lock
├── pyproject.toml
└── README.md

Credits

• YOLOv8 by Ultralytics
• Big help from https://github.com/TeogopK/Playing-Cards-Object-Detection