Real-time analysis system for Clash Royale gameplay using computer vision and machine learning.

• Live Card Detection: Real-time identification of cards using Roboflow API
• Game State Tracking: Opponent elixir count and deck monitoring
• Next Card Prediction: Opponent next card available prediction
• Web Interface: React frontend with live data visualization
• Non-Maximum Suppression: Prevents duplicate/overlapping detections
• Match Analytics: Historical data and performance statistics

• Python 3.8+
• Roboflow API key
• Iphone Screen Capture Software
• pip

cd clash-vision
pip install -r requirements.txt

cd clash-vision
python -m venv .venv
source .venv/bin/activate # On Windows: .venv\Scripts\activate

# 0. Make sure venv activated
# 1. Connect iPhone and open QuickTime Player (or OBS)
# 2. File > New Movie Recording > Select iPhone
# 3. Run region selector to define capture area
cd \Clash-Quant\clash-vision\src
# if using multiple displays change --monitor 1 to specify which display (1,2,..)
python -m clash_vision.inference.screen_preview --monitor 1
# Press 'r' to select iPhone screen area
# Drag out box to identify screen area
# Press 's' then 'esc' to save
# Press 'Ctrl+C' to exit 

# Basic game state tracking and model detection
# 0. Make sure venv activated
cd Clash-Quant\clash-vision.venv\Scripts\activate
python -m clash_vision.inference.live_inference --conf 0.05 --api_interval 0.5 --fps 30

# With web server for frontend access
python -m clash_vision.inference.live_inference --conf 0.05 --server-port 8080

# High performance mode
python -m clash_vision.inference.live_inference --conf 0.03 --api_interval 0.5 --fps 30

cd ../frontend
npm install
npm run dev  # Development server on http://localhost:3000

• GET /api/gamestate - Current game state and statistics
• GET /api/gamestate/detections - Active card detections
• POST /api/gamestate/reset - Reset match state
• GET /api/health - Health check

• Elixir Tracking: Real-time opponent elixir calculation
• Deck Discovery: Automatic opponent deck identification
• Card Cycling: Hand rotation and queue prediction
• Play History: Chronological card usage tracking
• Prediction Engine: Multi-factor next card forecasting

• ESC: Exit application
• R: Reset game state
• S: Save current frame (in screen preview mode)

No detections: Check confidence threshold and API key High CPU usage: Increase API interval or reduce FPS Capture issues: Ensure QuickTime window is visible and region is selected Web server errors: Check port availability and CORS settings

• Use --api_interval 0.5 for competitive analysis
• Set --nms-threshold 0.2 for stricter duplicate removal
• Enable --debug mode to analyze detection quality
• Monitor NMS filtering statistics in overlay
• Detection Model only trained on 8 cards so far
• Detection Model updates very slowly as of right now