Method, accuracy and sources

Capture

• A scan uses two still photos: front and right side. A back photo is optional.
• Each shot uses 5 consecutive frames, and the median position of each landmark is kept.
• A shot is rejected if the phone tilts more than 2° or the whole body, head to heel, is not in the frame.
• The member stands against a plain background in clothing that follows the body.

On the device

• Stance Scan finds 33 body landmarks with MediaPipe Pose Landmarker from Google (Apache-2.0).
• Detection runs on the phone or tablet. Photos are not uploaded and not stored.
• The result saved is a set of numbers: angles, bands, score, estimates and a confidence flag. No image, no face and no landmark coordinates are saved.

Why bands instead of exact degrees

• Camera-based joint angles agree with a goniometer within about ±8–11° (limits of agreement, Okuno 2026).
• Repeat measurements with MediaPipe show reliability (ICC) between 0.67 and 0.95 depending on the angle (Roggio 2024).
• An error of that size makes a single degree meaningless, so Stance shows bands and the change since the last scan.
• A value within 2° of a band line is marked “near the line” and should be read together with the next scan.

The 0–100 score

• The score starts at 100. Each item in the slight band subtracts half of its weight; each item in the notable band subtracts its full weight.
• Weights: head position 25, upper-body lean 20, shoulder height 15, pelvis height 15, knee direction 15, head tilt 10.
• The posture type combines front–back (balanced, head forward and rounding, leaning back) with left–right (even, left higher, right higher): 9 types.

The two-colour body figure

• Each posture type maps to muscle groups that tend to tighten and groups that tend to under-work.
• The rules follow the Janda approach to muscle balance (Page, Frank and Lardner, 2010).
• The figure uses two words only: “tends to tighten” and “tends to under-work”. It does not say a muscle is damaged or weak.

Circumference and body fat are estimates

• Neck, waist and hip circumference come from the body silhouette: width from the front photo, depth from the side photo, scaled by the height the member enters, then calculated as an ellipse (Ramanujan’s perimeter formula).
• Expected error is about ±2–3 cm. The upper guide is 25.6 mm, the result of a comparable single-camera method in Choudhary 2023. Our own pilot figures will replace this guide.
• Body fat is estimated with the Relative Fat Mass formula (RFM, Woolcott and Bergman 2018, checked against DXA in 12,581 adults). The US Navy formula (Hodgdon and Beckett 1984) is shown as a second value.
• If the two formulas differ by more than 5 points, the result is labelled “wide range”. Validation of RFM in Asian populations has not been confirmed.
• A muscle-mass reference value uses the Lee 2000 equation. If the member has a body-composition device reading, Stance shows that reading and hides its own estimate.
• Body fat and muscle estimates are shown for adults (18+) only, and every estimate carries the label “estimate”.

Not a medical device

Stance Scan is posture and body-shape guidance for exercise. It is not a medical device. The same statement is fixed in Japanese and Thai: