syed-baqir-kazmi/fire_smoke

🔥 RTSP CCTV Fire & Smoke Detection System

An enterprise-grade Fire & Smoke Detection web service for live CCTV RTSP streams, featuring automatic cloud snapshot uploads and browser-native web streaming for overall surveillance system integration.

Designed for seamless deployment to Hugging Face Spaces (Docker SDK) or local dockerized surveillance environments.

🚀 Features

• —Live RTSP Stream Transcoding for Web Browsers: Decodes RTSP feeds and transcodes them into high-speed browser-compatible MJPEG streams (GET /video_feed) viewable in any modern web browser or mobile client!
• —YOLO Deep Learning Inference: Pre-trained on fire and smoke object detection.
• —Temporal Hysteresis Confirmation:
• —Fire: 5 enter frames, 15 exit frames.
• —Smoke: 5 enter frames, 15 exit frames.
• —Automated Cloud Snapshot Upload: Automatically uploads hazard snapshots to cloud storage when incidents are detected.
• —Surveillance System REST & Streaming API:
• —GET /: Interactive Live Browser Surveillance Dashboard & Web Video Stream
• —GET /status: Current system detection state, FPS, bounding box coordinates, and metrics
• —GET /video_feed: High-speed MJPEG stream with live detection overlays for web integration
• —GET /latest_frame.jpg: Latest annotated frame image snapshot
• —GET /incidents: Recent incident history log with cloud image URLs
• —POST /detect: Synchronous image upload endpoint for on-demand fire/smoke inspection
• —POST /config: Dynamic RTSP stream URL and threshold updates on the fly

📁 Project Structure

rtsp_huggingface/
├── app.py                # FastAPI web server & browser dashboard
├── rtsp_detector.py      # Threaded RTSP stream engine & YOLO detector
├── config.py             # System configuration reader
├── requirements.txt      # Hugging Face deployment dependencies
├── Dockerfile            # Production Docker container setup for HF Spaces (Port 7860)
├── .env                  # Environment configuration file
├── .env.example          # Environment template for HF Space Secrets
├── README.md             # HF Space metadata & documentation
└── models/               # Model weights directory
    ├── fire.pt           # PyTorch YOLO model weights
    └── fire_openvino_model/

📡 Overall Surveillance System API Endpoints

1. Fetch System Status

• —Endpoint: GET /status

2. Live Web Video Feed (Browser Native MJPEG)

• —Endpoint: GET /video_feed
• —Usage: Embed directly in standard HTML <img> elements:

<img src="https://syed-baqir-kazmi-fire-smoke.hf.space/video_feed" width="100%">

3. Get Latest Snapshot Image

• —Endpoint: GET /latest_frame.jpg

4. Fetch Incident History

• —Endpoint: GET /incidents

5. Instant Image Inspection (REST Inference)

• —Endpoint: POST /detect