CivicSafe 🛡️

A real-time AI-powered public safety system that detects crowd overcrowding and fire hazards simultaneously using a live webcam feed — and triggers instant audio alerts.

---

🔍 What It Does

CivicSafe monitors live video and raises alerts for two types of safety risks:

Detection	Method	Alert Trigger
👥 Crowd	YOLOv8 person detection	When crowd count ≥ threshold
🔥 Fire	OpenCV HSV color masking	When fire-colored pixels > 4000

Both run in real time on a single webcam feed with a shared audio alert system (with 5-second cooldown to avoid repeated sounds).

---

🧠 How It Works

Live Webcam Feed
      │
      ├──► YOLOv8 (Person Detection)
      │         └── Rolling 10-frame average → Crowd Alert
      │
      └──► OpenCV HSV Masking
                └── Pixel threshold check → Fire Alert
                          │
                    🔔 Audio Alert (alert.wav) with cooldown

• Smooth counting: a 10-frame rolling buffer prevents flickering crowd counts from noisy detections
• Confidence filtering: only detections with ≥ 40% confidence are counted
• Modular code: civic_safe.py holds all logic; app_civic.py is the clean entry point

---

🛠️ Tech Stack

• Python 3.x
• YOLOv8 (Ultralytics) — real-time object detection
• OpenCV — video capture and HSV fire detection
• NumPy — array operations

---

🚀 Getting Started

1. Clone the repo

git clone https://github.com/Chandrika987/CivicSafe.git
cd CivicSafe

2. Install dependencies

pip install -r requirements.txt

3. Run

python app_civic.py

Press q to quit the live window.

---

⚙️ Configuration

In app_civic.py, adjust the crowd threshold:

crowd_detection(threshold=5)  # Alert when 5+ people detected

---

📁 Project Structure

CivicSafe/
├── civic_safe.py       # Core detection logic (crowd + fire + alerts)
├── app_civic.py        # Entry point
├── alert.wav           # Alert sound file
├── yolov8n.pt          # Pre-trained YOLOv8 nano model
├── requirements.txt
└── README.md

---

🔮 Planned Improvements (Ongoing)

• Streamlit web dashboard for live monitoring
• Hugging Face model integration for smarter scene classification
• Multi-camera support
• Logging alerts to a database with timestamps
• Deploy as a portable desktop app

---

👩‍💻 Author

Chandrika Pala — B.Tech CSE, SRKR Engineering College
LinkedIn · GitHub