[AfA2026-PhysicalAI] NetraHz

Arduino IDE

GoogleColab

Python programming language

Arduino® UNO™ Q 2GB

C1001 60 GHz mm Wave Indoor Fall Detection Sensor

Arduino® Cloud

<html><body><h3>1. Project Description</h3>NetraHz is a non-intrusive human monitoring system designed to address the reliability challenges in elderly and solo-living care. Our specific focus is solving "The Long Lie"—the dangerous period when a person falls or suffers a medical emergency and cannot call for help. Unlike traditional fall detectors that rely on simple binary triggers or invasive cameras that compromise privacy, NetraHz utilizes 60GHz mmWave radar to build an explainable, safety-critical architecture. The system is engineered as a Dual-Subsystem Architecture that separates long-term behavioral intelligence from immediate safety logic. This ensures that the system remains reliable, deterministic, and capable of understanding complex human behaviors without producing false positives. By "seeing" without seeing, NetraHz protects user privacy while ensuring continuous safety. <h3>2. Hardware Lineup</h3>Our system leverages the heterogeneous architecture of the **Arduino UNO Q** to process high-frequency radar data locally. Arduino UNO Q : The dual-core brain of the operation. The STM32 (Cortex-M33) core handles real-time sensor data acquisition and safety-critical reflexes, while the **Qualcomm (Cortex-A53)** Linux core runs the heavy statistical risk modeling. C1001 60GHz mmWave Radar: A high-precision FMCW radar sensor that detects human presence, motion intensity, and micro-Doppler vital signs (breathing) without capturing optical images. Power Supply: Standard 15W USB-C PD adapter to support the board and radar peaks. <h3>3. User Interface &amp; Feedback</h3>NetraHz is designed to be invisible to the user until it is needed. Privacy-First Interface : There are no cameras or wearables. The user simply lives their life, and the system monitors from the wall. Explainable Status Outputs: Unlike black-box AI, our system provides clear, tiered feedback states: Normal :The Fall Risk Index (FRI) is stable. Elevated Risk: FRI &gt; Threshold (indicates deteriorating mobility trends; suggests a check-up). Confirmed Emergency: Immediate fall detected by the safety subsystem. Alerting: Critical alerts are escalated via the board's connectivity options (Wi-Fi) to a caregiver dashboard, while local feedback can be managed via the board's built-in LEDs/Buzzer. <h3>4. The AI Model</h3>Instead of a pre-trained black-box classifier, NetraHz uses a **Streaming Behavioral Deviation Model** running on the Linux core of the Arduino UNO Q. Model Type: Unsupervised Statistical Anomaly Detection. Layer 2 (Adaptive Baselining): The system learns the individual's "Normal" by maintaining dynamic short-term (3-day) and long-term (30-day) baselines using Robust Statistics (Median &amp; Median Absolute Deviation). Layer 3 (Trend Analysis): It quantifies "Risk" not by generic thresholds, but by calculating the **Normalized Deviation** of the user's current behavior (Mobility, Activity, Transitions) against their own history. Local Processing: All learning and inference happen 100% on-device, ensuring no sensitive behavioral data leaves the home. <h3>5. Software Architecture</h3>The software is split into two logically independent subsystems that share sensor data but operate with distinct responsibilities: Subsystem 1: Layered Decision Funnel (Primary Intelligence) Running on the Linux Core, this is a general-purpose, 6-layer pipeline responsible for understanding long-term behavior. It aggregates raw radar noise into 1-hour windows and processes it through a strict funnel (Observation Baselining Context Modeling) to calculate a **Fall Risk Index (FRI)** bounded between 0 and 1. Subsystem 2: Deterministic Fall Logic (Parallel Safety) Running on the Real-Time Core (STM32), this is a standalone, non-AI subsystem. It runs a mandatory logic sequence (Radar Trigger Stationary State Activity Drop Persistence) to provide an immediate, conservative confirmation of emergency events. This ensures that even if the AI model is "thinking" or collecting data, the safety reflex remains active. <h3>6. Visuals &amp; Media</h3> <figure><img src="https://projects.arduinocontent.cc/eae16733-2a62-4a69-8dad-f7538fa1b391.jpg"/><figcaption>ARDUINO UNO Q (right) , C1001 mmWave RADAR (left)</figcaption></figure> <figure><img src="https://projects.arduinocontent.cc/412c7984-04fd-4df2-8da7-01af7bef7577.jpg"/><figcaption></figcaption></figure> YOUTUBE LINK: <a href="https://youtube.com/shorts/OFdqIIBCq7M?feature=share">https://youtube.com/shorts/OFdqIIBCq7M?feature=share</a><h3>Key Constraints</h3>The radar signal alone is never sufficient to trigger the flag. Any detection of recovery motion immediately cancels the sequence. This subsystem is purely deterministic; it does not use AI or probabilistic models. </body></html>

[AfA2026-PhysicalAI] NetraHz

A dual-core safety system powered by Arduino UNO Q that uses 60GHz mmWave radar to invisibly detect falls and track long-term health trends without cameras.