Pool Monitoring System

Pool monitoring system with object detection using YOLOv8.

Installation

pip install -r requirements.txt

Structure

detect_pool.py           # Main application
modules/
  config_manager.py      # Configuration management
  pool_boundary.py       # Pool area definition
  point_selector.py      # Interactive point selection
  object_detector.py     # YOLOv8 detection
config.json              # Application configuration
pool_config.json         # Pool coordinates (auto-generated)

Usage

1. Set video path in config.json
2. Run: python detect_pool.py
3. First run - select pool boundary points:
   • Click to add points
   • Drag to move
   • Right-click to remove
   • Enter to save (minimum 3 points)
4. Next runs - press 'c' to reconfigure or any key to use saved config

Keys during monitoring:

• q - Quit
• d - Toggle detection

Configuration

All settings in config.json. No default values - all parameters are required.

Key parameters

{
  "video": {
    "path": "pool_video.mp4"
  },
  "object_detection": {
    "enabled": true,
    "model_size": "n",
    "confidence_threshold": 0.15,
    "person_confidence_threshold": 0.5,
    "alert_delay_seconds": 2.0,
    "grace_period_seconds": 5.0,
    "skip_frames": 1,
    "process_size": 640
  }
}

Timing:

• alert_delay_seconds - Seconds in zone before alert
• grace_period_seconds - Alert persistence after last detection

Detection:

• confidence_threshold - Minimum confidence for general detection
• person_confidence_threshold - Confidence to classify as person (higher = fewer false positives)
• skip_frames - Process every N frames (higher = faster)
• process_size - Processed frame width (smaller = faster)

How it Works

Detects objects and people using YOLOv8, checks if detection center is inside pool boundary, triggers alerts based on zone occupation time.

• Detection enters zone → timer starts
• Stays for alert_delay_seconds → alert triggers
• Object leaves zone → alert persists for grace_period_seconds

Each tracked object maintains consistent classification (person vs object) throughout its lifetime to prevent alternation.