Ring Compass: Touchless Spatial Safety & Biometric Radar

Transforms ambient home cameras into a touchless radar for seniors: detects disorientation loops and falls, measures pulse through the air, and illuminates pathway lighting with zero wearables.

Comment

Inspiration

In our local neighborhood and among our own extended families, we witnessed the heartbreaking and exhausting reality faced by families caring for aging relatives with Alzheimer’s and progressive dementia. As evening turns into night, many seniors experience "sundowning syndrome"—an acute state of nocturnal confusion, disorientation, and anxiety.

They wake up at 2:00 AM in pitch-black corridors, pace back and forth repetitively in disoriented loops, or attempt to wander out through the front door into the dark.

When talking to elderly neighbors and family caregivers in our community, three devastating realities stood out:

Wearables are abandoned: Seniors with cognitive decline routinely take off, misplace, or forget to charge smartwatches and panic buttons before bed. They find them uncomfortable or confusing to operate. Binary motion sensors cry wolf: Standard motion sensors trigger blaring false alarms for harmless nighttime bathroom trips, disturbing sleep and exacerbating delirium. The catastrophic danger of falls: According to the CDC, over 1 in 3 adults aged 65+ fall each year, leading to fatal hip fractures and traumatic brain injuries. When an elder falls in the dark and cannot reach a phone, hours can pass before discovery. We asked ourselves: Why can't standard ambient optical cameras already installed in hallways and doorways act as a protective, touchless guardian—measuring vital signs through the air, detecting dangerous disorientation loops, and de-escalating nighttime panic without violating bedroom privacy or requiring wearable gadgets?

That question inspired us to build Elder Safety Mesh.

What it does

Elder Safety Mesh transforms standard ambient optical cameras into an autonomous, touchless spatial safety and biometric triage system for community eldercare:

Contactless Pulse & Respiration Extraction: Using Eulerian Video Magnification (EVM), the system measures heart rate (BPM) and breathing frequency through the camera lens by detecting sub-visual micro-luminosity shifts in the skin’s green optical channel ($\lambda = 530\text{ nm}$). Zero wearables, zero contact, zero battery maintenance.

Disorientation Cadence Index (DCI) Tracking: A 2D spatial centroid engine tracks movement trajectories. While purposeful walking follows a linear path ($DCI \approx 1.0\text{x}$), agitated pacing loops cause the trajectory-to-displacement ratio to spike ($DCI \ge 3.0\text{x}$), detecting disorientation before a crisis escalates into a fall.

Deterministic Fall Impact Detection: Combines vertical centroid acceleration ($a_y \ge 2.8g$) and bounding aspect-ratio collapse to immediately distinguish accidental mechanical falls from sitting or resting.

Gentle Ambient De-Escalation: Instead of blaring loud panic sirens that aggravate delirium, the system autonomously illuminates warm amber pathway lighting ($\approx 590\text{ nm}$, which preserves nighttime melatonin) and sounds a soothing 432 Hz acoustic grounding tone on local smart speakers.

Instant Emergency Dispatch: If an elder collapses or wanders out a perimeter doorway, an automated forensic emergency alert with verified vital signs and timestamps is dispatched to family caregivers and emergency contacts.

100% Volatile RAM Privacy Guarantee: Video frames are processed in-memory and discarded in milliseconds. Zero raw video or audio ever leaves the home or touches the cloud.

How we built it

Elder Safety Mesh was engineered from the ground up as a zero-cloud-cost, local-first edge architecture:

  1. Dual-Camera Edge Perception Pipeline OpenCV 4.10 & NumPy: Ingests dual 60 FPS video streams (Hallway Corridor & Front Porch Doorway) simultaneously in volatile edge RAM. 2D Spatial Centroid Kinetics: Calculates continuous coordinate matrices across a sliding 30-frame temporal window to derive: $$\text{DCI} = \frac{L}{\Delta D} = \frac{\sum_{i=1}^{N} \sqrt{(x_i - x_{i-1})^2 + (y_i - y_{i-1})^2}}{\sqrt{(x_N - x_0)^2 + (y_N - y_0)^2}}$$

  2. Eulerian Video Magnification (EVM) Isolates forehead Region-of-Interest (ROI) skin pixels and isolates the green optical absorption band ($530\text{ nm}$). Employs a 4th-order temporal Butterworth bandpass filter ($0.75\text{ Hz} \le f \le 2.5\text{ Hz}$, corresponding to $45 - 150\text{ BPM}$): $$H(s) = \frac{G_0 \cdot \omega_0^4}{\prod_{k=1}^{4} (s - p_k)}$$ Extracts live heart rate and respiratory waveforms in real time via Fast Fourier Transform (FFT).

  3. Open Model Context Protocol (MCP) Server Implemented a self-hosted Model Context Protocol (MCP) server via FastAPI and JSON-RPC 2.0 (mcp.json). Exposes standardized edge tools: camera_stream_ingest, evm_extract_vitals, toggle_pathway_light, and trigger_chime.

  4. Physical Smart Home Hardware Node (ESP32) An onboard ESP32 microcontroller connected to the local network acts as the physical de-escalation bridge: GPIO 4: Drives high-luminosity Amber LED strips illuminating floor pathways. GPIO 15: Voice-coil acoustic transducer emitting a harmonic 432 Hz de-escalation chime.

  5. Responsive Monitoring Console Built with React 18, Vite, TypeScript, and Tailwind CSS. Features a 5-tab real-time HUD with live 2D phase space plots, Recharts vital trendlines, and system audit logs.

Challenges we ran into

Optical Micro-Luminosity in Dim Night Light: Extracting blood volume pulses through camera lenses under low ambient lighting was challenging due to sensor noise. We solved this by implementing temporal averaging across forehead pixel masks and applying a narrowband Butterworth filter that rejects sensor flicker while isolating arterial pulses.

Preventing False Alarms from Normal Night Bathroom Trips: An elder walking directly to the bathroom shouldn't trigger an emergency alert. We developed the Disorientation Cadence Index (DCI) to mathematically differentiate purposeful linear walking ($DCI \approx 1.0\text{x} - 1.4\text{x}$) from repetitive, anxious wandering circles ($DCI \ge 3.0\text{x}$).

Sample-Accurate Edge Latency: To achieve medical-grade triage, the system had to analyze video, extract vitals, and actuate physical smart lighting in under 50 milliseconds. By processing frames in multithreaded Python memory buffers, we achieved an end-to-end decision latency of $38.4\text{ ms}$.

Accomplishments that we're proud of

Zero Wearables Needed: Protecting seniors without forcing them to wear battery-powered watches or bracelets to bed.

Complete Zero-Cloud-Cost Privacy: Building a system where 100% of video frames exist only in volatile RAM, completely eliminating privacy risks and recurring cloud subscription fees.

Sub-42ms Reaction Speed: Successfully triggering physical amber pathway lighting and acoustic grounding in under 42 milliseconds from the moment erratic pacing is identified.

Fully Working Physical Hardware Node: Proving the end-to-end loop by actuating physical LEDs and speakers through an open-source ESP32 microcontroller bridge.

What we learned

Digital Signal Processing for Healthcare: Deepened our mastery of Fourier transforms, Butterworth bandpass filtering, and Eulerian video magnification for non-contact medical telemetry.

Agentic Protocol Standards: Learned how to architect modular tool-calling systems using the open Model Context Protocol (MCP).

Empathy-Driven Engineering: Discovered that in eldercare technology, loud sirens and invasive cameras cause more harm than good—true safety must be ambient, gentle, and respectful of personal dignity.

What's next for Elder Safety Mesh: Touchless Spatial Safety & Biometrics

Neighborhood Mutual-Aid Circle: Expanding alert routing so if a solo-living elder wanders outside at 2:00 AM, trusted, pre-approved neighbors receive a discreet notification to check on them safely.

Micro-Doppler Radar Integration: Pairing optical feeds with $60\text{ GHz}$ mmWave radar to monitor respiration through thick blankets without line-of-sight.

Community Care-Home Pilot: Deploying Elder Safety Mesh edge nodes in local assisted living facilities to reduce night-shift nurse burnout and eliminate preventable nocturnal fall injuries.

Built With

Share this project:

Updates

Submission history