Intelligent Human-Zone Tracking using Computer Vision

A real-time safety monitoring system for hazardous industrial zones that detects people, verifies safety-vest compliance, and tracks dwell-time in danger areas—raising instant alerts when safety rules are violated.

---

🚀 Features

• Human Detection
  Frame-by-frame person detection powered by YOLOv8 for high speed and accuracy.

• Safety-Vest Classification
  Color-based vest presence check:

  • ❌ No Vest → Immediate RED alert
  • 🟡 Yellow Vest → Start zone-timer & issue YELLOW warning if threshold exceeded

• Zone-Based Timer
  Tracks how long each person remains in the danger zone; configurable time limits trigger audible/visual warnings.

• (Optional) Face Recognition
  Integrate dlib-based face matching to tag and log individual IDs in the dashboard.

• Web Dashboard
  Flask-powered UI to stream video, display alerts, timers, and safety logs in real time.

---

📦 Getting Started

Prerequisites

• Python 3.8+
• CUDA-enabled GPU (recommended for real-time performance)
• Roboflow Dataset: Safety Vests

Installation

1. Clone this repository

   git clone https://github.com/your-username/human-zone-tracking.git
   cd human-zone-tracking
   
   

2. Create & activate a virtual environment

   python3 -m venv venv
   source venv/bin/activate     # Linux/macOS
   venv\Scripts\activate.bat    # Windows

3. Install dependencies

   pip install -r requirements.txt

---

⚙️ Usage

1. Download or link your dataset

   • Export your annotated images from Roboflow (YOLO format).
   • Place train/, val/, test/ folders under data/safety_vests/.

2. Train YOLOv8 model (optional)

   yolo detect train model=yolov8n.yaml data=data/safety_vests/ epochs=50 imgsz=640

3. Run inference & dashboard

   python app.py \
     --weights runs/train/exp/weights/best.pt \
     --source 0 \
     --zone-file config/zone.json \
     --time-threshold 10

   • --source can be a webcam index or video file.
   • --zone-file defines polygon(s) of hazardous area(s).
   • --time-threshold is the max seconds allowed in-zone with a vest.

---

📊 Results & Demo

• Detection accuracy: mAP₅₀ > 92%
• Real-time performance: ≈25 FPS on RTX 2060
• Demo Video: YouTube Link

Example Detections

Here’s a sample frame with our human-zone overlays:

And here’s the “blue square” test pattern we used for calibration:

🛠️ Project Structure

├── app.py                 # Main Flask app + inference loop  
├── models/                # YOLOv8 model configs & weights  
├── data/                  # Train/Val/Test datasets  
│   └── safety_vests/      
├── utils/
│   ├── detector.py        # YOLOv8 wrapper
│   ├── vest_classifier.py # Color-based vest check
│   ├── timer.py           # Zone dwell-time logic
│   └── face_recog.py      # (Optional) dlib integration
├── requirements.txt
└── README.md

---

📝 Configuration

• config/zone.json

  {
    "zone_polygon": [[x1,y1], [x2,y2], …, [xn,yn]]
  }

• config/settings.yaml

  time_threshold: 10       # seconds
  alert_colors:
    no_vest: "red"
    vest:    "yellow"

---

🤝 Contributing

1. Fork the repo
2. Create a feature branch (git checkout -b feature/YourFeature)
3. Commit your changes (git commit -m "Add YourFeature")
4. Push to branch (git push origin feature/YourFeature)
5. Open a Pull Request

Please adhere to the existing coding style and include tests for new functionality.

---

👥 Team & Acknowledgments

• Rasul & Zhanibek – YOLOv8 human & vest detection
• Olzhas & Bakkustar – Face-recognition module
• Ilyas – Zone timer & alert logic

Special thanks to Roboflow Universe Dataset for annotated safety-vest images.

---