🚀 Real-time Crowd Counting Detection System

AI-powered bidirectional people counting and occupancy monitoring with YOLO11

---

📖 Overview

A production-ready AI system for real-time people counting and occupancy monitoring using state-of-the-art YOLO11 object detection. Perfect for hallways, entrances, retail spaces, and smart building management.

Key Capabilities:

• 🎯 High-Speed Detection: Optimized YOLO11n model.
• ⚡ Intel NCS2 Acceleration: Full support for OpenVINO FP16 acceleration.
• 👥 Real-Time Counting: Accurate "IN" and "OUT" tracking.
• 🍓 Raspberry Pi Native: Built for Pi 4/5 with picamera2 support.
• 📊 Ultra-Compact UI: Minimalist dashboard with "OUT" and "NOW" metrics.
• ⚙️ Auto-Start: Systemd service for 24/7 autonomous operation.

---

🚀 Quick Start

1. Installation on Raspberry Pi

We provide a single script to set up everything (Python, OpenCV, OpenVINO, Service).

# Clone the repository
git clone https://github.com/ayyoubbenmansour/Crowd_Counting_Detection_RaspberryPi.git
cd Crowd_Counting_Detection_RaspberryPi

# Run the auto-setup script
chmod +x setup_rpi.sh
./setup_rpi.sh

2. Run the Application

You can run it manually or as a background service.

Manual Start (for testing):

python3 app.py

Access the dashboard at http://<YOUR-PI-IP>:5000

Background Service (Auto-start on boot):

sudo systemctl start crowd_counting.service

---

⚡ Intel NCS2 Acceleration

To get maximum FPS on a Raspberry Pi 4, we recommend using the Intel Neural Compute Stick 2.

1. Export the Model:

yolo export model=yolo11n.pt format=openvino half=True

2. Run with Acceleration:

python3 crowd_counting_rpi.py --model yolo11n_openvino_model/ --interactive

See NCS2_GUIDE.md for full details.

---

📊 Dashboard Metrics

The UI is designed to be unobtrusive overlaying the video feed.

Metric	Description
OUT	Total number of people who have walked OUT of the zone.
NOW	Current Occupancy (people currently inside the zone).
Zone Status	Color-coded status (Green=OK, Orange=Warn, Red=Critical).

---

💻 Advanced Usage

Custom Line Setup

Run in interactive mode to click-and-draw your counting zone.

python3 crowd_counting_rpi.py --interactive

Headless Mode (SSH)

Run without a GUI window, perfect for remote servers.

python3 crowd_counting_rpi.py --no-display --output debug_session.mp4

Optimize for Speed

Reduce resolution to gain FPS.

python3 crowd_counting_rpi.py --resolution 640 360

---

📂 Project Structure

Crowd_Counting_Detection_RaspberryPi/ ├── app.py # Flask Web Application (Video Streaming) ├── crowd_counting_rpi.py # Core AI Logic & Processing ├── setup_rpi.sh # One-click Installation Script ├── crowd_counting.service # Systemd Auto-start Service ├── DEPLOYMENT.md # Detailed Pi Deployment Guide ├── NCS2_GUIDE.md # Intel NCS2 Acceleration Guide ├── TRAINING_GUIDE.md # Guide to Fine-Tune Model ├── DATASETS.md # List of Public Datasets ├── logs.json # Local Log Storage (Generated at runtime) ├── templates/ # Web Interface Templates (layout, index, settings, logs) ├── yolo11n.pt # YOLO Model Weights └── README.md # Documentation

---

## 🤝 Contributing

Contributions welcome!
- 🐛 Report bugs via Issues
- 💡 Suggest features
- 🔧 Submit pull requests

---

<div align="center">

**© 2025/2026 Real-time Crowd Counting**

</div>