Rehabit

Rehabit turns your webcam into a rehab coach—guided drills, live symmetry/ROM & posture scores, and a Gemini-powered report you can share with your clinician.

Comment

Inspiration

What it does

Inspiration

Rehab after stroke or injury is hard to access and even harder to measure at home. We wanted something anyone could run in a browser that gives objective feedback, not just “how did it feel?”

What it does

Rehabit turns a webcam into a home rehab coach. Users follow short guided drills—arms up, elbow flex/extend to shoulders, and sit-to-stand. We track joint angles and posture to compute:

  • Symmetry Score (left vs right)
  • Range of Motion (ROM) %
  • Rep quality (speed, control, trunk lean)

A Gemini call turns these numbers into a readable plain-English report and simple progress notes you can share. No wearables, no install.

How we built it

  • Pose & CV: MediaPipe Pose + OpenCV + NumPy for keypoints, angles, smoothing (Savitzky–Golay + outlier clipping).
  • App: Flask + Flask-SocketIO backend; lightweight JS frontend for camera preview, overlays, timers.
  • Analytics: Rep segmentation, max/min angle per rep, symmetry & ROM calculations, CSV/JSON export.
  • Summaries: Gemini API receives a structured session JSON and returns a readable report for users/clinicians.

Challenges

Lighting/occlusion causing pose jitter, calibrating angles across body sizes, counting reps reliably, and packaging it so it runs smoothly in-browser.

Accomplishments

An end-to-end flow: guided drills → real-time overlays → symmetry/ROM scores → auto-generated report. We built clear exercise animations and handle one-side-weaker cases.

What we learned

Designing stable metrics (not noisy ones), smoothing pose data, writing prompts for medical-adjacent language, and UX that motivates without overwhelming.

What’s next

We’re committed to take Rehabit to the next level. In the coming weeks we will:

  • Add personalized baselines, more upper-limb drills, and side-view capture.
  • Launch a clinician dashboard (cohort views, PDF exports, alerts for large left–right gaps).
  • Keep data privacy-first (on-device where possible, encrypted storage, consent logs).
  • Start a commercial pilot with a partner clinic, validate against goniometer measures, add FHIR/EHR export, and align with HIPAA/PIPEDA + RTM reimbursement.

If we had more time this weekend: we’d ship auto-calibration for camera distance, richer coaching cues (“slow the right side”), polished PDF reports, clinician invites, and additional drills (reach, grasp, gait).
We’re not diagnosing—we make progress visible and actionable, and we plan to keep working on this beyond the hackathon.

Built With

Share this project:

Updates