Open-source · local-first · GPL-3.0

Open Gait

Real-time running form analysis from a side, front, or back camera — pose landmarks to biomechanics, streamed to a local dashboard.

View on GitHub Run the demo

Camera in. Form cues out.

A Rust pipeline captures frames, estimates BlazePose landmarks, and computes gait metrics. Results leave your machine only if you choose — the dashboard talks to a local WebSocket.

01 Capture Live camera or a scripted synthetic demo reel
02 Pose Detector ROI + 33-point landmark model (ONNX)
03 Biomechanics Cadence, lean, overstride, hip drop, valgus, crossover
04 Dashboard Live overlay, trends, recording, and PDF debrief

Three camera planes

Switch viewpoint from the UI — the same session can analyse sagittal form and frontal-plane cues without restarting the backend.

Side

Sagittal

Cadence, knee flexion at contact, torso lean, overstride.
Front

Anterior

Hip drop, lateral trunk lean, knee valgus, crossover gait.
Back

Posterior

Same frontal metrics from behind — useful on a treadmill belt.

Privacy by design

Analysis runs on your computer. Turn Video off in the dashboard to stop JPEG frames over the WebSocket — skeleton and metrics keep updating.

Try the demo in two terminals

No camera or ONNX models required. The backend loops a 70-second gait reel with changing form scenarios — good for exploring the UI or filming a walkthrough.

# Terminal 1 — backend demo reel
make demo

# Terminal 2 — Open Gait Dashboard
cd web && npm install && npm run dev

Then open http://localhost:5173. Full live-camera setup and flags live in the README.