Inspiration
Pulse oximeters, ECG monitors, even a working thermometer are things many of us take for granted. But hundreds of millions of people in rural and low-income regions have none of that at home. The only "device" almost everyone already owns is a smartphone. That gap — a phone in every pocket, but no way to use it to check your own vitals — is what VitaLens is built to close.
What it does
VitaLens turns a smartphone's front camera into a contactless vital-sign screener. Point the camera at your face for 30–45 seconds, and it estimates your heart rate, respiration rate, and an SpO₂ proxy using remote photoplethysmography (rPPG) — detecting tiny, invisible color changes in your skin caused by blood flow. An AI triage layer then turns those numbers into a plain-language recommendation: self-monitor, consult a doctor, or seek urgent care.
How I'm building it
This is a Round 1 idea submission, so there's no working prototype yet — but the technical approach is scoped out: a Progressive Web App (React) using getUserMedia to access the camera with no app install needed, Google MediaPipe Face Mesh and OpenCV for facial tracking and signal extraction, CHROM/POS rPPG algorithms with bandpass filtering to isolate the pulse waveform from noise and lighting flicker, and a lightweight model plus an LLM-based assistant to turn the waveform into HR/RR/SpO₂ estimates and plain-language guidance.
Challenges I anticipate
The biggest ones are accuracy under real-world conditions: variable lighting, motion during capture, and known skin-tone bias in existing rPPG research. I'm planning to address these with adaptive lighting normalization, a motion-detection guard that prompts re-capture, and calibration against a skin-tone-diverse dataset — while being upfront that VitaLens is a screening aid, not a diagnostic device.
What's next for VitaLens
If shortlisted for the Final Round, the next step is a working browser-based prototype: real-time camera capture, live rPPG signal extraction, and a first pass at the AI triage layer — then testing accuracy against a pulse oximeter as ground truth.
Log in or sign up for Devpost to join the conversation.