RTSP Blackmist Detector 🚨💨

A Dockerized system that watches an RTSP video stream for fume / dark smoke using a LOCAL vision‑language model, stores results in InfluxDB, and exposes them through a lightweight web dashboard.

---

✨ Features

• 📡 RTSP live‑stream frame extraction
• 🤖 AI analysis using an Ollama VLM (recommended: Qwen2.5-VL)
• 📈 Data storage in InfluxDB v1
• 🌐 Web interface to view detections and history
• 🔒 HTTPS support via your own cert.pem and key.pem

Web page example:

---

🧩 Requirements

Before using this project, you must install/configure the following:

1. Docker

Install Docker for your OS:
https://docs.docker.com/get-docker/

2. InfluxDB v1

Install InfluxDB v1.x and note your credentials Fill the file accordingly

3. Ollama (⚠ Required for AI analysis)

• Must be installed on the host machine
• Must be accessible from inside Docker → listen on 0.0.0.0
• Install from: https://ollama.ai

To expose Ollama:

ollama serve --host 0.0.0.0

4. Download the model

Recommended model:

ollama pull qwen2.5vl

5. Clone the repository and open it

gh repo clone Coldness00/Fumes-Detector
cd Fumes-Detector

6. Provide certificates

These are required for HTTPS access. At the root of the repo, put:

cert.pem
key.pem

This command line will create them for 10years.

openssl req -x509 -newkey rsa:2048 -keyout key.pem -out cert.pem -days 3650 -nodes

---

⚙️ Configuration (docker-compose.yml)

You must edit docker-compose.yml before running the stack.

Below are fields you MUST understand and adapt:

---

❌ Do NOT change

FOLDER_PATH=/app/images
DB_PATH=/app/data/processed_images.db

---

✅ Must be customized

BASE_URL

Machine’s local IP + port
Example:

http://192.168.1.15:9822

EXTERNAL_URL

Use only if routing through a reverse proxy.
Otherwise: leave it commented.

REFRESH_TIME

Time between image captures (seconds).

• Lower = faster detection
• Higher load on CPU/GPU

PROMPT

Recommended to keep as provided.
You can tweak, but preserve formatting.

OLLAMA settings

Defaults are safe, but can be adjusted if needed.

CAMERA_NAME

Friendly camera display name.

RTSP_URL

Your RTSP address including ID and password if required.

InfluxDB block

Adjust host, port, database, username, password.

TZ

Timezone (example):

TZ=Europe/Paris

---

📁 Volumes

You must map local folders to the container:

./images:/app/images
./data:/app/data

These store:

• extracted frames
• processed database

---

🔌 Ports

Adjust if running multiple instances.

---

🚀 Start the project

Run the stack:

docker compose up -d

Check logs:

docker compose logs -f

---

🖥️ Access the Web Dashboard

Open:

http://<BASE_URL>:<PORT> -> http://192.168.1.15:9822
https://<BASE_URL>:<PORT>

If using self‑signed certificates, your browser may show a warning—this is expected.

---

🗂 Project Structure

.
├── data/                     <- local DB and persistent data
├── images/                   <- extracted frames
├── main.py                   <- main application
├── db.py                     <- database handler
├── docker-compose.yml
├── Dockerfile
├── cert.pem / key.pem
└── requirements.txt

👨‍💻 Grafana Dashboard

You can find enclosed an example of a Grafana Dashboard, copy/paste as a new dashboard, then edit the entites name accorgindlgy.

---

❓ Questions / Issues

Feel free to open an Issue or request enhancements.
Contributions and suggestions are always welcome! 🚀