Not analyzed.
Mapping method, formulas & limits
Method v1.0.0: body pose + face mesh + expression coefficients. Yellow points: estimated face surface; green lines: visible body joints.
Joint angle: acos((A−B)·(C−B) / (|A−B||C−B|)), converted to degrees, using aspect-corrected image coordinates. Normalized position: (point − shoulder midpoint) / shoulder width. Motion speed: displacement / elapsed seconds / frame diagonal × 100. Coverage: accepted frames / sampled frames × 100.
Mouth opening: lip gap / mouth width. Eye opening: eyelid gap / eye width. Brow height and chin/cheek proportions use face dimensions for scale. Neck-axis tilt uses ear midpoint relative to shoulder midpoint; it is not a clinical neck angle.
Body landmarks require visibility ≥ 0.6 and must be inside the frame. Face mesh has no individual landmark visibility gate; occluded face points may be inferred. Facial coefficients are model estimates (not emotions or biometric identity). Approximately two samples/sec can miss blinks and fast motions.
Not measured: individual teeth, tongue, hair strands, neck skin surface, detailed fingers, physical inch measurements or audio-lip accuracy. No smoothing or retargeting is applied. Camera angle/occlusion affect all measurements.
Face model documentation · Body model documentationOnly visible, confident landmarks are measured. A chest-up reference cannot supply reliable leg motion. Arms/wrists and facial landmarks are measured; detailed fingers require separate analysis. These readings are not yet converted into an animation model’s conditioning format.