How on-device pose processing works

ErgoVision uses Google’s MediaPipe Pose Landmarker, running natively on your phone’s CPU/GPU, to detect 33 body landmarks per frame at roughly 15 frames/second.

From those landmarks we compute joint angles — trunk flexion, neck angle, upper arm abduction, knee bend, etc. — and feed them into the validated REBA and RULA scoring tables.

This architecture means:

• No internet required for assessment capture or scoring.
• Worker video never leaves the device. Only numeric scores and (if you save them) thumbnail images sync to the cloud.
• Inference is fast. A 3-minute recording processes in under 15 seconds on modern phones.