Professional thermal intelligence dashboard for firefighter operations. Displays normal (RGB) and thermal camera feeds side-by-side with live AI analysis: person detection, heat zones with direction and distance, flashover risk scoring, and an AI-powered tactical advisor. Video sources are read from local disk only; no upload.

1. Login — Access is gated by username/password; after login you see a sidebar of up to 7 firefighter camera feeds.
2. Feed selection — Each feed is a pair of aligned videos: normal (video N.mp4) and thermal (video N_thermal_intense.mp4). Selecting a feed loads both streams.
3. Dual-feed processing — For each frame pair:
   • RGB feed is used for person detection (YOLO) and depth estimation (MiDaS). The RGB frame is shown unchanged on one side.
   • Thermal feed is converted to grayscale, temporally smoothed, and analyzed for heat zones. Overlays (person boxes, heat zones, HUD) are drawn only on the thermal view.
4. Risk engine — A rolling window of average thermal intensity is used to compute a flashover risk score (0–100) and trend (RISING / STABLE / FALLING). HIGH risk triggers an on-screen evacuation warning.
5. Command Advisor — On demand, the current snapshot (victim count, risk, heat zones) is sent to a local LLM (Ollama) to generate short tactical recommendations for the Fire Captain.

• Python 3.10+
• pip
• Ollama (optional, for Command Advisor): install from ollama.com and run ollama pull gemma3:4b

Place aligned normal and thermal videos in the project’s video directory (default: fire fighter cam/).

Default single pair:

• normal.mp4 — RGB feed
• thermal.mp4 — Thermal feed (same length, aligned by timestamp)

Multi-feed (7 cameras):

• video 1.mp4 … video 7.mp4 — Normal feeds
• video 1_thermal_intense.mp4 … video 7_thermal_intense.mp4 — Thermal feeds

If your folder or filenames differ, set environment variables (see Configuration).

cd Fire
pip install -r requirements.txt

YOLOv8 nano (yolov8n.pt) and MiDaS weights download automatically on first run.

streamlit run app.py

Open the URL shown (e.g. http://localhost:8501). Log in (default credentials in app), pick a camera feed in the sidebar, and click Start stream. Both feeds run in sync; metrics and risk update from the thermal pipeline.

Optional environment variables:

Example (Unix/macOS):

export FIRE_CAM_VIDEO_DIR="/path/to/your/videos"
streamlit run app.py

A sample .env.example is included; copy to .env and load it with direnv or python-dotenv if you use one.

Fire/
├── app.py              # Streamlit dashboard (login, sidebar feeds, dual view, metrics, Building Info, Command Advisor)
├── config.py           # Video paths and validation (get_video_paths, validate_video_sources)
├── video_processor.py  # YOLOv8 + ByteTrack, MiDaS depth, heat zones, dual-feed processing, HUD
├── risk_engine.py      # Flashover risk score and trend from thermal intensity history
├── llm_advisor.py      # Ollama (Gemma 3 4B) tactical briefing from cached result and risk
├── utils.py            # Resize, thermal effect, drawing, direction/distance helpers
├── requirements.txt   # Python dependencies (streamlit, ultralytics, torch, opencv, etc.)
├── .env.example        # Optional env template
└── README.md           # This file

pip install -r requirements.txt
streamlit run app.py

streamlit run app.py --server.port 8501 --server.address 0.0.0.0

Example unit (adjust paths and user):

[Unit]
Description=Prometheus Thermal Intelligence Dashboard
After=network.target

[Service]
Type=simple
User=www-data
WorkingDirectory=/opt/prometheus
Environment="FIRE_CAM_VIDEO_DIR=/var/lib/prometheus/videos"
ExecStart=/usr/bin/streamlit run app.py --server.port 8501 --server.address 0.0.0.0
Restart=on-failure

[Install]
WantedBy=multi-user.target

Example Dockerfile:

FROM python:3.11-slim
WORKDIR /app
COPY requirements.txt .
RUN pip install --no-cache-dir -r requirements.txt
COPY . .
ENV FIRE_CAM_VIDEO_DIR=/data
VOLUME /data
EXPOSE 8501
CMD ["streamlit", "run", "app.py", "--server.port=8501", "--server.address=0.0.0.0"]

Build and run (mount the folder with your video pairs):

docker build -t prometheus .
docker run -p 8501:8501 -v "/path/to/videos:/data" prometheus

• Sync: RGB and thermal are advanced by frame index; use the same length and frame rate for correct alignment.
• Thermal input: Heat zones and flashover risk use the thermal feed (grayscale). Person detection and depth use the RGB feed; overlays are drawn on the thermal view.
• Risk levels: LOW (<40), MEDIUM (40–69), HIGH (70–100). HIGH triggers the on-screen evacuation warning.
• Command Advisor: Requires Ollama running with gemma3:4b. If Ollama is not available, the briefing button shows a connection error.

Built for BisonHacks. MIT License.