Casino Chip Detection System

Project Overview

This state-of-the-art chip detection and processing system is designed for real-time object detection, segmentation, and classification of gaming chips in casino environments. The system processes high-resolution video feeds from multiple cameras to detect and count gaming chips, identify their positions and assignments, and manage data through a Flask-based server with real-time updates via SocketIO.

Built with a modular architecture, the system leverages multiple specialized YOLO models for distinct tasks including chip extraction, counting, cage identification, and position tracking. The system is designed to be adaptable across multiple tables without table-specific references, making it scalable for casino-wide deployment.

Key Features & Capabilities

• Multi-model YOLO pipeline with specialized models for extraction, counting, cage assignment, and position detection
• Real-time 4K video processing from dual cameras with high-resolution 2048x2048 image processing
• Advanced chip categorization system distinguishing chip types and categories (cash vs. non-cash)
• Polygon-based annotation system for precise cage and position assignments using point-in-polygon tests
• Flask server with SocketIO for real-time client communication and RESTful API endpoints
• Interactive Tkinter GUI with scrollable displays for camera feeds, extracted objects, and chip values
• External API integration for centralized data storage with comprehensive hand data management
• Multi-threaded architecture ensuring GUI responsiveness during concurrent server operations

Technical Implementation

The system employs multiple YOLO models (likely YOLOv8 based on Ultralytics) optimized for real-time detection and segmentation. Specialized models handle distinct tasks: extractor model for object detection, counter model for chip counting, cage identification model for category assignment, and position identification model for spatial positioning on the table.

High-resolution processing at 2048x2048 pixels ensures detailed segmentation and accurate chip detection, with initial detection at 640x640 for speed optimization. The system uses configurable confidence thresholds (0.45-0.65) for different models, balancing detection reliability with processing speed. OpenCV with DirectShow backend and MJPG codec ensures efficient 4K video capture and processing.

Results & Performance Metrics

The system achieves real-time processing with 30ms GUI update intervals and maintains detailed chip inventories with dynamic updates. The enhanced implementation features advanced chip categorization capabilities, tracking counts and values for each chip type and category, suitable for complex casino environments with diverse chip sets.

The polygon-based annotation system ensures precise localization using point-in-polygon tests, with upscaled annotations adjusted to 2048x2048 resolution for consistency. External API integration enables seamless connectivity with management systems, providing comprehensive hand data including image paths, chip details, cage positions, timestamps, and batch codes for complete activity tracking.