Posture+ is a real-time posture monitoring system that runs entirely in your browser using your webcam. It detects posture instability through temporal analysis — identifying sustained slouch patterns rather than momentary dips.
No wearables. No sensors. No recordings. Just your laptop camera and five seconds to calibrate.
Most posture tools work like this:
if shoulder_angle > threshold:
alert()This misses the point. A single bad frame doesn't mean bad posture. Posture+ instead evaluates how posture behaves over time, detecting:
- Sustained forward head posture
- Lateral head tilt
- Shoulder imbalance
- Posture fatigue during long sessions
Alerts only fire when instability is genuinely sustained — not on random dips.
Posture+ is built on a research framework developed for real-time posture stability analysis. The work has been accepted for presentation at IEEE CSPA 2026, modeling posture as a temporal stability signal rather than a geometric threshold.
The mathematical formulation behind the scoring system is intentionally not disclosed in this repository.
The system scores posture using a composite metric called the Posture Stability Index (PSI).
PSI evaluates:
- Posture quality (deviation from baseline)
- Temporal stability (variance over time)
- Degradation patterns (worsening drift)
- Recovery behavior (time to recover from red zones)
- Fatigue signals (session-level decline)
Instead of binary good/bad alerts, PSI provides a continuous 0–100 stability score throughout a session.
flowchart TD
A[Webcam Stream] --> B[MediaPipe Pose]
B --> C[Landmark Extraction]
C --> D[Feature Engine]
D --> E[Temporal Stability Engine]
E --> F[PSI Computation]
F --> G[Real-Time UI]
F --> H[Session Analytics]
F --> I[Alert Engine]
H --> J[Supabase Database]
G --> K[Dashboard]
G --> L[Monitor]
G --> M[Reports]
All processing runs locally in the browser. No video is uploaded anywhere.
| Feature | Description |
|---|---|
| Real-Time Monitoring | Tracks upper body landmarks every frame via MediaPipe Pose |
| Temporal Analysis | Evaluates posture across time windows, not single frames |
| PSI Score | Continuous 0–100 stability score with live chart |
| Smart Alerts | Configurable delay, repeat interval, sound, and browser notifications |
| Adaptive Recalibration | Baseline drifts only during verified stable posture — slouch can't become the new normal |
| Session Analytics | PSI timeline, zone exposure, fatigue detection, stability degradation index |
| Privacy First | No video stored or uploaded — only posture metrics |
| Guest Mode | No account needed — session vanishes on reload |
Frontend — React 19, Vite, TypeScript, Tailwind CSS
Computer Vision — MediaPipe Pose (runs in-browser via WASM)
Backend — Supabase, PostgreSQL, Row Level Security
Infrastructure — Vercel, Turborepo, pnpm workspaces
| Table | Purpose |
|---|---|
users |
Profile and auth metadata |
sessions |
Per-session analytics (PSI, duration, zone distribution, fatigue) |
posture_samples |
Raw posture samples captured during sessions |
daily_summaries |
Aggregated daily statistics for the dashboard chart |
# Clone
git clone https://github.com/DarkBytezz/PosturePlus.git
cd PosturePlus
# Install
pnpm install
# Start
pnpm devOpen http://localhost:5173 and allow camera access when prompted.
Environment variables — create .env in apps/web/:
VITE_SUPABASE_URL=your_supabase_url
VITE_SUPABASE_ANON_KEY=your_supabase_key
- Sit upright and face the screen directly
- Click Calibrate in the Monitor tab
- Hold your posture for ~5 seconds
This establishes your personal baseline. The system adapts to your natural seated position, not a generic template.
- Works best with laptop webcams and seated desk posture
- User should be centered and reasonably well-lit in frame
- Mobile layout is currently limited
- Mobile layout optimization
- Session drill-down into raw samples
- CSV export for research use
- Long-term trend analysis
- Posture coaching mode
Manan Verma B.E. Computer Science (AI & ML), Chandigarh University
MIT — see LICENSE
If you found this project interesting, consider leaving a ⭐