A safety-focused room occupancy monitoring system for industrial environments, leveraging Ultralytics YOLOv8 for person detection.

Repository: https://cnb.cool/MindDockAI/yolo-room-occupancy-safety

• Person Detection: Uses YOLOv8 to detect humans.
• Line Crossing Counting: Tracks entries and exits through defined virtual doors. (You must physically cross the line to be counted).
• Global Safety State: Maintains a global count of people in the room.
• Robot Control Logic:
  • STOP: If count > 0 OR System Error.
  • START: Only if count == 0.
• Fail-Safe:
  • Watchdog for camera frame drops.
  • Auto-reconnection logic for cameras.
  • Defaults to "Occupied" (Stop) on any failure.

• OS: Windows 11, macOS, or Linux.
• Python: 3.8 or higher.
• Hardware: Webcam or RTSP Camera (one per door).

git clone https://cnb.cool/MindDockAI/yolo-room-occupancy-safety.git
cd yolo-room-occupancy-safety

Open a terminal and run:

# Create a virtual environment
python3 -m venv venv

# Activate the environment
source venv/bin/activate

# Install dependencies
pip install -r requirements.txt

Open PowerShell or Command Prompt and run:

# Create a virtual environment
python -m venv venv

# Activate the environment (PowerShell)
.\venv\Scripts\Activate.ps1
# OR for Command Prompt:
# venv\Scripts\activate.bat

# Install dependencies
pip install -r requirements.txt

Note for macOS Users: If you encounter errors related to ultralytics or torch, ensure you have the latest version of pip: pip install --upgrade pip.

Edit config.yaml to match your physical setup.

cameras:
  - id: 0
    source: "0"          # Webcam ID (0) or RTSP URL ("rtsp://...")
    line_start: [100, 200]
    line_end: [500, 200] # Define the virtual door line (x, y)
    direction_in: "down" # Detection direction logic (currently vector based)

• source: Use 0, 1 for USB webcams, or an RTSP stream URL.
• system.debug_mode: Set to true to see the video feed with detection overlays.

Ensure your virtual environment is activated, then run:

python -m src.main

The system will start monitoring. Press q to exit if debug window is open.

You can test the logic without a camera by generating a mock video.

1. Generate the video:

   python tests/mock_generator.py
   # Creates mock_video.mp4
   

2. Update Config: In config.yaml, set source: "mock_video.mp4".
3. Run Main:

   python -m src.main
   

• Camera Watchdog: If a camera stops sending frames for >0.5s (configurable), the system enters Error State.
• Negative Count: If the count accidentally goes below 0 (logic error), it resets to 0 but logs a warning. In a strict safety version, this should trigger a Stop.
• Redundancy: Verify the logical "In/Out" direction matches your physical camera mounting orientation.

• YOLO Tracking Error (no module named lap): The project uses lapx instead of lap. Ensure you run pip install -r requirements.txt.
• Camera Timeout: If using a webcam that takes time to initialize, you may see immediate "Camera TIMEOUT" errors. Increase system.max_missing_frames in config.yaml (e.g., to 60 or higher).

MIT (or specify your license)