🛡️ GuardianAI

Intelligent Security Surveillance System using Computer Vision & AI

Real-time surveillance solution with Fire/Smoke detection, Stranger/Known person recognition, Fall detection,
Telegram alerts, GUI management and automatic event recording.

---

🎯 Overview

GuardianAI is an intelligent security surveillance system, optimized for 24/7 operation with high performance. The system uses advanced AI models to detect dangers and send instant alerts via Telegram.

✨ Key Features

Feature	Description
🔥 Fire & Smoke Detection	YOLO (OpenVINO/ONNX) with smart post-processing filters, minimizing False Positives
👤 Face Recognition	InsightFace + SORT Tracker, classifying Known/Stranger persons
🚨 Fall Detection	RTMPose + ONNX model, real-time pose analysis
📸 Infrared Camera (IR)	Auto-detection and appropriate filter switching
📱 Telegram Alerts	Send photos/videos, interactive confirmation buttons, periodic heartbeat
🤖 AI Assistant	Compatible with OpenAI API (Gemini/LM Studio/Ollama)
🎬 Auto Recording	Automatically record clips on events and send with alerts
🖥️ Management GUI	Beautiful interface with CustomTkinter

---

🚀 Installation

System Requirements

• OS: Windows 10/11 or Linux
• Python: 3.12
• RAM: Minimum 4GB (recommended 8GB+)
• Camera: Webcam, IP Camera (RTSP), or video file

Step 1: Clone repository

git clone https://github.com/your-username/GuardianAI.git
cd GuardianAI

Step 2: Create virtual environment (recommended)

python -m venv .venv
.venv\Scripts\activate  # Windows
# or
source .venv/bin/activate  # Linux/macOS

Step 3: Install dependencies

pip install -r requirements.txt

Step 4: Configuration

Edit file config/config.yaml:

# === Telegram Bot ===
telegram:
  token: ${TELEGRAM_TOKEN:""}
  chat_id: ${TELEGRAM_CHAT_ID:""}

# === Camera ===
camera:
  # Webcam: 0 | File: "video.mp4" | RTSP: "rtsp://user:pass@ip:554/stream"
  sources: 0
  target_fps: 10

# === AI Assistant (optional) ===
ai:
  enabled: false
  api_base: ${AI_API_BASE:"https://api.openai.com/v1"}
  api_key: ${AI_API_KEY:""}
  model: ${AI_MODEL:"gpt-4o-mini"}

💡 Tip: Use environment variables with syntax ${ENV:"default"} to secure sensitive information.

Step 5: Run the application

python main.py

---

📂 Project Structure

GuardianAI/
├── main.py                 # Entry point - Initialize and orchestrate system
├── config/                 
│   ├── config.yaml         # Main configuration file
│   └── settings.py         # Load config, supports ${ENV:"default"}
├── core/                   
│   ├── camera.py           # Camera class: read frames, IR detection, pipeline
│   ├── camera_manager.py   # Multi-camera management, processing threads
│   ├── recorder.py         # Alert video recording
│   └── detection/          
│       ├── fire.py         # FireDetector: YOLO + RGB/IR filters
│       ├── face.py         # FaceDetector: InsightFace
│       ├── person.py       # PersonTracker: YOLO + SORT + Face ID
│       └── fall.py         # FallDetector: RTMPose + ONNX
├── bot/                    
│   └── *.py                # Telegram Bot, AI Assistant
├── gui/                    
│   ├── app.py              # Main GUI Application
│   ├── panels/             # Panels (settings, cameras, faces...)
│   ├── widgets/            # Custom widgets
│   └── styles.py           # Theme and styling
├── utils/                  
│   └── *.py                # StateManager, SpamGuard, AlarmPlayer...
└── Data/                   
    └── Model/              # AI models (YOLO, InsightFace, ONNX...)

---

🔧 Feature Details

🔥 Fire & Smoke Detection

• Uses YOLO with OpenVINO/ONNX backend for optimal performance
• Smart filters to minimize False Positives:
  • Color analysis (HSV ranges)
  • Flicker detection
  • Reflection and bright light filtering
  • Texture analysis (entropy)
• Dedicated IR mode: Auto-switches filters when camera is in infrared mode

👤 Face Recognition

• InsightFace for high-quality face embeddings
• SORT Tracker for continuous multi-person tracking
• ReID (Re-Identification): Remembers people when they leave and return
• Management GUI: Add new people, build embeddings

🚨 Fall Detection

• RTMPose (from rtmlib) for 17-point pose estimation
• ONNX model classifies falls based on pose sequences
• Analyzes keypoint velocity and location
• Visual skeleton overlay display

📸 Infrared Camera (IR) Support

• Auto-detects IR mode every 10 frames
• Maintains 30-frame history for stable results
• Dedicated IR filters (brightness, variance, hot spot)
• Smoke detection disabled by default in IR mode

---

📱 Telegram Bot Usage

Available Commands

Command	Description
/start	Start bot, show instructions
/status	Check system status
/detect	Toggle detection mode
/alarm	Control alarm siren
/get_image	Capture image from camera

Alert Handling

When receiving alerts, you can interact via buttons:

• ✅ Confirm: This is a real event
• ❌ Dismiss: False positive
• 🔇 Mute: Stop alarm siren

---

⚙️ Advanced Configuration

See config/config.yaml to adjust:

• Camera: FPS, resolution, process interval
• Detection: Confidence thresholds, filter parameters
• Tracker: IOU threshold, max age
• Recorder: Duration, extension time
• Telegram: Response timeout, spam protection

---

📜 License

This project is released under the MIT License. See LICENSE file for more details.

Made with ❤️ by HuyHAP & Minh Tri