Real-Time Poker Card Detection on Jetson Orin

This project demonstrates real-time playing card detection using YOLO11s optimized with TensorRT FP16 on the NVIDIA Jetson Orin NX. The system accurately detects and classifies all 52 playing cards from a standard deck in real-time.

Hardware:

  • NVIDIA Jetson Orin NX 16GB

Software Environment:

  • JetPack 6.2 (L4T R36.4.7)
    CUDA 12.6
    TensorRT 10.7
    cuDNN 9.17
    Python 3.10
    PyTorch 2.8
    Ultralytics 8.3.40

Key Features

  • 52-Class Detection: Recognizes all cards in a standard deck (2-10, J, Q, K, A across all four suits)
  • TensorRT FP16 Optimization: Model exported to TensorRT engine with FP16 precision for maximum inference speed on Jetson
  • GStreamer Hardware Acceleration: Custom camera pipeline using GStreamer with NVIDIA hardware encoding/decoding for minimal CPU overhead

Github: GitHub - farukalamai/realtime-poker-card-detection-jetson: Real-time poker card detection on NVIDIA Jetson Orin NX 16 GB.