Pose Tracking

RepFi verifies your reps with AI pose tracking (also called pose estimation, skeleton tracking or body tracking). The camera detects key points of your body — head, shoulders, elbows, hips, knees and more — and connects them into a skeleton that moves with you.

The model we use: MediaPipe Pose

RepFi uses MediaPipe Pose by Google AI Edge.

Why	Detail
Runs in the browser	No server inference needed — works on the user’s device
Real time	Smooth tracking even during fast movement
33 body points	Enough detail to judge push-up form
Scales	Each live room runs tracking locally, so 10 rooms cost us nothing extra

Why not the alternatives

MoveNet Thunder — very accurate with less jitter, but heavier and without a ready live-overlay pipeline for our use case.
Rokoko Vision — excellent for character animation / motion capture, not for a real-time web bounty service.

For “overlay a skeleton on a webcam and count push-ups in real time”, MediaPipe Pose is the best fit.

How a rep is verified

The camera frame is sent to MediaPipe Pose on your device.
The model returns 33 landmarks (the skeleton).
RepFi measures the elbow angle and body alignment.
A rep counts when you go from up → down → up with correct form.
On each valid rep, your balance increases by $0.01 instantly.

Privacy

Pose detection runs locally in your browser. The skeleton and rep count are what drive your rewards. Your live video is shown to viewers only while you are in a room.

Requirements

A working camera
Good lighting
Your full body in frame — camera to the side or facing you (phone on the floor works)