AI Guard Demo with Arduino UNO Q, Modulino Sensors, and Local NPU Face Recognition

Windows ARM64 laptop with Snapdragon X Elite or compatible NPU

Arduino IDE

Arduino® UNO™ Q 4GB

USB WebCam

Modulino™ Buzzer

Qwiic Cables Bundle

Onnx Runtime

Arduino App Lab

Modulino™ Distance

<html><body>The Arduino UNO Q is a great way to connect physical sensing with local AI. In this project, the board listens to a Modulino Distance sensor and uses a Modulino Buzzer as the physical output. A Windows ARM64 laptop runs the heavier computer vision work locally on its NPU. The demo works like this: when somebody comes close to the distance sensor, the Arduino UNO Q reports the distance to the laptop. The laptop captures a frame from the browser camera, detects the face with a MediaPipe ONNX detector, and then runs CavaFace through ONNX Runtime QNN to create a face embedding. That embedding is compared with a local known-face database. If the person is recognized, the dashboard shows a trusted match and the buzzer stays quiet. If the person is not recognized, the dashboard shows an unknown face alert and the Arduino UNO Q activates the Modulino Buzzer. The goal is to make edge AI visible. Instead of showing only code or terminal logs, the project shows the full path from a real-world sensor event to a local AI decision and then back to a physical hardware response. Project Hub Section: Step-By-Step Tutorial Prerequisites You need: <pre>```text </pre><pre>Arduino UNO Q </pre><pre>Modulino Distance </pre><pre>Modulino Buzzer </pre><pre>Qwiic cables </pre><pre>USB cable </pre><pre>Windows ARM64 laptop with Snapdragon X Elite or compatible NPU </pre><pre>Laptop camera or USB webcam </pre><pre>Arduino IDE </pre><pre>Python 3 for Windows ARM64 </pre><pre>``` </pre> You also need the local ONNX model assets: <pre>```text </pre><pre>cavaface.onnx </pre><pre>media_pipe.onnx </pre><pre>``` </pre> The GitHub repository intentionally does not include those model files because they are runtime assets. Keep them local unless you have confirmed the license allows redistribution. Step 1: Clone The Demo Repository Open PowerShell and clone the project: <pre>```powershell </pre><pre>cd C:\Users\Public\Downloads\arduino </pre><pre>git clone https://github.com/shivaylamba/cavaface-detection-arduino-unoq-npu.git </pre><pre>cd C:\Users\Public\Downloads\arduino\cavaface-detection-arduino-unoq-npu </pre><pre>``` </pre> Step 2: Create The Python Environment Use native Windows ARM64 Python so the NPU runtime can load correctly. <pre>	 </pre><pre>```powershell </pre><pre>python -c "import platform; print(platform.machine())" </pre><pre>``` </pre> Expected output: <pre>```text </pre><pre>ARM64 </pre><pre>``` </pre> Create the virtual environment and install the demo dependencies: <pre>```powershell </pre><pre>python -m venv laptop_ai_guard\.venv </pre><pre>.\laptop_ai_guard\.venv\Scripts\Activate.ps1 </pre><pre>python -m pip install --upgrade pip </pre><pre>python -m pip install -r laptop_ai_guard\requirements-windows-npu.txt </pre><pre>``` </pre> Step 3: Add The CavaFace And MediaPipe Models Create the model folders: <pre>```powershell </pre><pre>New-Item -ItemType Directory -Force laptop_ai_guard\models\cavaface </pre><pre>New-Item -ItemType Directory -Force laptop_ai_guard\models\media_pipe </pre><pre>``` </pre> Copy the model files into these exact paths: <pre>```text </pre><pre>laptop_ai_guard\models\cavaface\cavaface.onnx </pre><pre>laptop_ai_guard\models\media_pipe\media_pipe.onnx </pre><pre>``` </pre> The app uses the MediaPipe model to find faces and the CavaFace model to create 512-dimensional face embeddings. ### Step 4: Upload The Arduino UNO Q Firmware Open Arduino IDE and load this sketch: <pre>```text </pre><pre>firmware\arduino_q_face_guard\arduino_q_face_guard.ino </pre><pre>``` </pre> Select the Arduino UNO Q board and the correct port, then upload the sketch. Step 5: Start The Local AI Demo Run the demo app from the repository root: <pre>```powershell </pre><pre>.\laptop_ai_guard\.venv\Scripts\python.exe -u laptop_ai_guard\run_guard.py ` </pre><pre>--hardware-source routerbridge ` </pre><pre>--camera-source browser ` </pre><pre>--browser-timeout 180 ` </pre><pre>--face-detector mediapipe ` </pre><pre>--model-runtime onnx-qnn ` </pre><pre>--threshold 0.50 ` </pre><pre>--proximity-threshold-mm 700 ` </pre><pre>--trigger-cooldown 2 ` </pre><pre>--poll-interval 0.25 </pre><pre>``` </pre> If the app cannot find the Arduino ADB tool automatically, pass the path: <pre>```powershell </pre><pre>.\laptop_ai_guard\.venv\Scripts\python.exe -u laptop_ai_guard\run_guard.py ` </pre><pre>--hardware-source routerbridge ` </pre><pre>--adb-path "%LOCALAPPDATA%\Arduino15\packages\arduino\tools\adb\32.0.0\adb.exe" ` </pre><pre>--camera-source browser ` </pre><pre>--browser-timeout 180 ` </pre><pre>--face-detector mediapipe ` </pre><pre>--model-runtime onnx-qnn ` </pre><pre>--threshold 0.50 ` </pre><pre>--proximity-threshold-mm 700 ` </pre><pre>--trigger-cooldown 2 ` </pre><pre>--poll-interval 0.25 </pre><pre>``` </pre> When the app starts, it opens a local browser dashboard: <pre>```text </pre><pre><a href="http://127.0.0.1:8765/">http://127.0.0.1:8765/ </a> </pre><pre>``` </pre> Allow camera access and keep the browser tab open. Step 6: Check The Live Demo Screen The `Live Demo` screen shows the main demonstration: <pre>```text </pre><pre>Camera preview </pre><pre>Distance sensor reading </pre><pre>Face match result </pre><pre>Match confidence </pre><pre>Buzzer decision </pre><pre>Sensor-to-AI-to-hardware flow </pre><pre>``` </pre> Move your hand toward the Modulino Distance sensor. When the distance goes below the configured threshold, the app runs face detection and face matching. Step 7: Add A Known Face Open the `Add Known Face` screen in the dashboard. Enter a demo name, choose the number of samples, face the camera, and start the capture. The app stores CavaFace embeddings in: <pre>```text </pre><pre>laptop_ai_guard\known_faces\embeddings.npz </pre><pre>``` </pre> This does not retrain CavaFace. The model stays fixed. The app simply adds new local face templates to the known-face database. Step 8: View The Known Faces Database Open the `Known Faces` screen. It shows the names currently stored in the local database and the number of embeddings saved for each name. This makes the demo easier to explain because the audience can see who the app is expected to recognize before the physical interaction starts. Step 9: Run The Physical Demo Now run the full interaction: 1. Open the Live Demo screen. 2. Keep the camera tab open. 3. Bring a hand or object close to the Modulino Distance sensor. 4. Watch the dashboard show that the sensor triggered recognition. 5. Let the app run MediaPipe face detection and CavaFace matching locally. 6. If the face is known, the dashboard shows a match and the buzzer remains off. 7. If the face is unknown, the dashboard shows an alert and the buzzer sounds. </body></html>

Arduino UNO Q Face Guard Demo Example

AI Guard Demo with Arduino UNO Q, Modulino Sensors, and Local NPU Face Recognition

Build a proximity-triggered face recognition demo where Arduino UNO Q senses a person nearby and a laptop NPU runs local AI before driving a physical buzzer response.