Drone Detection — RUBIK Pi 3

Roni Bandini · May 2026 · Argentina · @RoniBandini

Audio-based drone detection running on a Qualcomm RUBIK Pi 3. An Edge Impulse audio classifier listens continuously for drone sounds; when confidence exceeds a configurable threshold, a webcam snapshot is taken and analyzed by a local instance of OpenClaw to confirm the presence of a drone, identify its make and model, and decide whether to log or alert. Alerts are delivered via Telegram and a physical LED on the GPIO.

---

How It Works

Microphone
    │
    ▼
Edge Impulse AudioImpulseRunner (.eim model)
    │  drone confidence %
    ▼
Threshold check  ──── below threshold ──▶  meter display (no action)
    │
    │  above threshold + cooldown elapsed
    ▼
fswebcam snapshot
    │
    ▼
OpenClaw vision VLM
    │  JSON: action / drone_type / size / reason
    ▼
┌─────────────────────────────────────┐
│  ignore  │  log (CSV)  │  alert     │
│          │  LED on     │  LED on    │
│          │             │  Telegram  │
└─────────────────────────────────────┘

---

Hardware

Component	Details
Board	Qualcomm RUBIK Pi 3
OS	Ubuntu (aarch64)
Microphone	USB microphone on default ALSA device
Camera	USB webcam on /dev/video0
LED	Connected to GPIO pin 571

---

Requirements

System packages

sudo apt install fswebcam python3-pip libportaudio2 libportaudiocpp0 portaudio19-dev

Python packages

pip3 install edge_impulse_linux -i https://pypi.python.org/simple
pip3 install requests pyaudio opencv-python python-periphery

OpenClaw

Install and configure OpenClaw with a vision-capable model.

Edge Impulse model

Download your trained .eim model from Edge Impulse Studio.

---

Configuration

All settings are at the top of drone.py:

Variable	Default	Description
confidenceThreshold	75.0	Audio confidence % required to trigger visual check
visualCooldownSeconds	360	Minimum seconds between camera captures
gpioPin	571	GPIO pin number for the alert LED
ledOnTime	3	Seconds the LED stays on per alert
telegramToken	""	Telegram Bot API token
telegramChatId	""	Destination Telegram chat or channel ID
openClawAgent	"main"	OpenClaw agent name
webcamDevice	"/dev/video0"	USB webcam device path
csvLogPath	"drone.csv"	Local CSV detection log

---

Usage

Directly

python3 drone.py /path/to/model.eim
python3 drone.py /path/to/model.eim --verbose

---

Output

Console — confidence meter

Drone: [────────────────        ]  42.0%
Drone: [████████████████████████]  87.3% 🚨 THRESHOLD EXCEEDED

Console — visual analysis block

--- OPENCLAW VISUAL ANALYSIS ---
Drone visible : ✅ True
Model         : DJI Mini 4 Pro
Size          : small
Action        : 🚨 alert
Reasoning     : A small white quadcopter with a gimbal camera is visible flying at low altitude.
--------------------------------

CSV log (drone.csv)

timestamp	audio_confidence_pct	drone_type	action	snapshot_path
2026-05-10 14:32:01	88.4	DJI Mini 4 Pro	alert	/home/ubuntu/pic_20260510_143201.jpg

Telegram alert

🚨 ALERT - Drone detected
Model: DJI Mini 4 Pro
Size: small
Confidence: 88.40%
Reason: Small quadcopter visible at low altitude near perimeter.

---

Demo

https://www.youtube.com/watch?v=togAkkwYE-E

---

Complete tutorial

https://docs.edgeimpulse.com/projects/expert-network/multilayer-ai-agent-drone-detection-rubik-pi-3

---

Project Structure

├── drone.py      # Main detection script
├── drone.csv       # Detection log (auto-created on first run)
└── README.md # This file

---

License

MIT License — see LICENSE for details.