HOW POSE SCORING WORKS:
INSIDE THE MATH & FORMULAS.
Pose Scoring in POSEHANUM calculates the cosine similarity and angular difference between 3D vector triplets formed by anatomical keypoints (such as shoulder-elbow-wrist or hip-knee-ankle). Deviations between live camera coordinates and reference matrices are normalized, weighted by joint visibility confidence, and aggregated into a real-time 0% to 100% alignment score at 60 FPS.
1. Extracting Joint Angle Triplets
Rather than relying on absolute pixel positions (which vary with distance and screen resolution), POSEHANUM evaluates scale-invariant relative joint angles. For any three connected keypoints $A$, $B$, and $C$ (where $B$ is the vertex joint, e.g. the elbow):
u_vec = Point_A - Point_B, v_vec = Point_C - Point_B
cos(θ) = (u_vec · v_vec) / (||u_vec|| * ||v_vec||)
θ = arccos(clamp(cos(θ), -1.0, 1.0)) * (180 / π)
2. The Normalized Scoring Equation
The aggregate score compares N critical joint angles between the live user posture and the target reference:
Score = 100 * (1.0 - (Σ w_i * (|θ_user_i - θ_target_i| / 180°)) / (Σ w_i))
Where $w_i$ represents the visibility confidence score provided by the neural model. If a limb is occluded behind the torso, its weight decreases automatically to prevent false penalties.
3. Dynamic Visual Thresholds & Auto Shutter Lock
Lime Green (Locked)
Triggers 2-second hold countdown for auto capture.
Cyan (Minor Shift)
Whispers targeted audio prompt (e.g. "tilt chin 5°").
Orange (Realign)
Guides baseline stance and body direction.