What AI does
Google’s pretrained MediaPipe Pose Landmarker Lite estimates body landmarks. The site draws visible points and calculates 2D elbow and knee angles from their positions.
AI × Lifestyle / Interactive experiment
Movement Lab
See a frame through a pose model. Then check what it missed.
Your file stays on this device. The model downloads from this site on first use (about 17 MB). No camera access or account is needed.
Personal archive · upper-body exampleObservation, not a verdict
What does the model see?
Start with the sample or choose your own file. For video, pause or scrub to a frame and analyse it. A full-body side view helps when looking at knees.
How to read the result
My contribution
The browser workflow, local worker integration, confidence filtering, geometry and bilingual explanation. The pretrained model is Google’s work; I have not trained or clinically validated it.
Where it stops
Camera angle, occlusion and cropping change the estimate. This is a single-frame exploration, not a biomechanical assessment, training recommendation or medical tool. It does not recognise football tricks.