Safety Monitor v1.3.0 🏭

Real-time industrial safety monitoring system using computer vision. Detects safety violations, PPE compliance, falls, and restricted zone intrusions in live video streams.

Features

• 🎥 Real-time Video Processing - Webcam, IP cameras, RTSP streams, video files
• 👤 Person Detection - YOLOv8-based detection with confidence scoring
• 🦺 PPE Detection - Helmet, vest, gloves, masks, goggles
• 📍 Zone Management - Custom restricted areas with intrusion alerts
• 👥 Fall Detection - Automatic fall detection using motion & aspect ratio analysis
• 🔔 Smart Alerts - Configurable alerts with cooldown periods & logging
• 📹 Video Recording - Auto-record violations
• 📊 Live Statistics - FPS counter, frame count, violation tracking
• 🎨 Visual Annotations - Real-time frame annotations with status indicators

Tech Stack

• Python 3.8+
• OpenCV - Video processing
• PyTorch - Deep learning
• YOLOv8 - Object detection
• Streamlit - Dashboard
• Loguru - Logging

Installation

1. Clone Repository

git clone https://github.com/Ayesha037/safety_monitor.git
cd safety_monitor

2. Virtual Environment

# Windows
python -m venv venv
venv\Scripts\activate

# Linux/Mac
python -m venv venv
source venv/bin/activate

3. Install Dependencies

pip install -r requirements.txt

4. GPU Support (Optional)

# For NVIDIA GPU (replace cu118 with your CUDA version)
pip install torch torchvision --index-url https://download.pytorch.org/whl/cu118

Quick Start

Basic Usage

# Webcam (default)
python main.py

# IP Camera / RTSP Stream
python main.py --source "rtsp://user:pass@192.168.1.100:554/stream"

# Video File
python main.py --source "video.mp4"

# Headless Mode (no display)
python main.py --no-display

# Enable Recording
python main.py --record

Keyboard Controls

Key	Action
Q	Quit
S	Screenshot

Configuration

Edit config/config.yaml:

system:
  name: "Safety Monitor"
  version: "1.3.0"
  log_level: "INFO"

video:
  source: "0"                    # 0=webcam, URL, or video path
  width: 1280
  height: 720
  fps_target: 30
  resize_factor: 1.0             # 0.5 = 50% resolution (faster)
  frame_skip: 0                  # Skip frames for speed (0=process all)

model:
  detector: "models/yolov8n.pt"
  device: "cuda"                 # cuda or cpu
  confidence_threshold: 0.5
  nms_iou_threshold: 0.45
  max_detections: 100

ppe:
  enabled: true
  person_class_id: 0
  ppe_model_path: "models/ppe_classifier.pt"
  required_ppe: ["helmet", "vest"]
  confidence_threshold: 0.7

fall_detection:
  enabled: true
  aspect_ratio_threshold: 0.5
  velocity_frames: 5
  velocity_threshold: 50
  min_area_fraction: 0.01

zones:
  enabled: true
  zones:
    - name: "Restricted Area"
      type: "polygon"
      points: [[100, 100], [500, 100], [500, 400], [100, 400]]
      color: [0, 0, 255]
      alert_on_intrusion: true

alerts:
  enabled: true
  cooldown_seconds: 5
  log_path: "logs/alerts.log"

recording:
  enabled: true
  output_path: "logs/videos/violations.mp4"
  codec: "mp4v"

display:
  show_fps: true

Project Structure

safety_monitor/
├── config/                 # Configuration management
├── models/                 # Detection & classification models
│   ├── detector.py            # YOLOv8 object detector
│   ├── ppe_classifier.py      # PPE detection
│   ├── fall_detector.py       # Fall detection engine
│   └── safety_pipeline.py     # Main processing pipeline
├── utils/                  # Utility modules
│   ├── video_capture.py       # Video input handling
│   ├── zone_manager.py        # Zone management & intrusion
│   ├── alert_manager.py       # Alert system
│   ├── drawing.py             # Frame annotation utilities
│   └── logger.py              # Logging setup
├── dashboard/              # Streamlit dashboard (optional)
├── logs/                   # Output: alerts, screenshots, videos
├── main.py                 # Main entry point
├── requirements.txt        # Python dependencies
└── yolov8n.pt             # Pre-trained YOLOv8 model

Core Components

ObjectDetector

Detects persons in frames using YOLOv8.

detector = ObjectDetector(
    model_path="models/yolov8n.pt",
    device="cuda",
    conf_threshold=0.5
)
detections = detector.detect(frame)

PPEClassifier

Classifies PPE items for each detected person.

ppe = PPEClassifier(
    model_path="models/ppe_classifier.pt",
    conf_threshold=0.7
)
ppe_status = ppe.classify(frame, detections)

FallDetector

Detects falls using motion analysis.

fall_detector = FallDetector(
    aspect_ratio_threshold=0.5,
    velocity_frames=5
)
is_falling = fall_detector.detect(detection, prev_detection)

SafetyPipeline

Orchestrates all detectors and returns comprehensive results.

pipeline = SafetyPipeline(
    detector=detector,
    ppe=ppe,
    fall=fall_detector,
    zone_mgr=zone_mgr,
    alert_mgr=alert_mgr
)
result = pipeline.process(frame)

Performance Tips

Optimize for Speed

video:
  resize_factor: 0.5      # Process at 50% resolution
  frame_skip: 2           # Process every 3rd frame
model:
  confidence_threshold: 0.4  # Lower = more detections (slower)

Optimize for Accuracy

video:
  resize_factor: 1.0      # Full resolution
  frame_skip: 0           # Process all frames
model:
  confidence_threshold: 0.7  # Higher = fewer false positives

Troubleshooting

CUDA Out of Memory

python main.py --resize-factor 0.5
# Or switch to CPU
export TORCH_DEVICE=cpu
python main.py

No Detections

• Lower confidence threshold in config.yaml
• Ensure camera/video source is working
• Check lighting conditions

Low FPS

• Reduce resolution: resize_factor: 0.5
• Skip frames: frame_skip: 2
• Use CPU-only if GPU memory is limited

Camera Not Found

# Test camera
python -c "import cv2; cap = cv2.VideoCapture(0); print(cap.isOpened())"

Output

Console

=== Safety Monitor v1.3.0 starting ===
System ready. Press Q to quit, S to screenshot.
[FRAME 0001] Persons: 3 | Violations: 1 | Zone Intrusions: 0 | FPS: 28.5
[ALERT] PPE Violation: Person #2 missing helmet
[ALERT] Zone intrusion in Restricted Area
Screenshot saved: logs/screenshot_1686754323.jpg

Logs

• logs/safety_monitor.log - Main system log
• logs/alerts.log - Alert events
• logs/screenshots/ - Captured frames
• logs/videos/ - Recorded violations

Deployment

Docker

FROM nvidia/cuda:11.8.0-runtime-ubuntu22.04
WORKDIR /app
COPY requirements.txt .
RUN pip install -r requirements.txt
COPY . .
CMD ["python", "main.py"]

docker build -t safety-monitor .
docker run --gpus all -it safety-monitor

Linux Service

# Copy service file
sudo cp safety_monitor.service /etc/systemd/system/
sudo systemctl daemon-reload
sudo systemctl enable safety_monitor
sudo systemctl start safety_monitor

# Check status
sudo systemctl status safety_monitor
journalctl -u safety_monitor -f

Contributing

1. Fork the repository
2. Create feature branch: git checkout -b feature/new-feature
3. Commit changes: git commit -m 'Add new feature'
4. Push to branch: git push origin feature/new-feature
5. Open Pull Request

Roadmap

• Web-based dashboard
• REST API
• Multi-camera support
• Custom model training
• Mobile app

License

MIT License - See LICENSE file for details

Support

• 📋 Issues: GitHub Issues
• 💬 Discussions: GitHub Discussions

Acknowledgements

• OpenCV - Computer vision library
• Ultralytics - YOLOv8 framework
• PyTorch - Deep learning framework

---

⭐ If this project helped you, please star the repository!

Made with ❤️ for workplace safety