Everything documented in one place: the invention, the computer-vision methodology, the exact scoring formula, comparison against human judges, the audit trail from the 2026 championship, the appeals process, and the patent status. Built for scientific scrutiny.
Classical Yogasana scoring depends on human panels visually estimating alignment, hold duration and difficulty. This has three long-standing problems: viewing angle (a judge on the left cannot see spinal alignment as accurately as a judge in front), fatigue (attention drifts across a 6-hour tournament), and inter-judge variance (two experienced judges routinely score the same performance 0.5–1.5 points apart). The SEGA AI Judge System was invented by Prof Segar to standardise all three — not to replace human expertise, but to give every athlete the same measurement tool applied identically, and every judge a common reference layer.
1080p 30fps front-facing camera + optional 45° side camera. Two-camera synchronisation via NTP timestamp.
Real-time human pose estimation extracts 33 body landmarks per frame (head, shoulders, elbows, wrists, hips, knees, ankles, feet, hands, fingertips).
Sub-pixel refinement on critical joints — spinal midline, sacral tilt, cervical curve — using SEGA's proprietary Yogasana-specific model.
Each landmark set compared against the ideal geometry for the declared asana, indexed by the UYSF Tariff Table.
Frame-to-frame variance of landmark positions during the hold window, quantified as CoG micro-tremor.
Three sub-scores (Execution, Stability, Difficulty) combined with the tariff multiplier to produce the final 0.00–10.00 score.
Athlete video streamed to the scoring station. Frame extraction at 30fps.
Skeleton tracker outputs 33 landmarks × N frames.
Geometric & temporal analysis. Execution + Stability + Difficulty sub-scores.
Final score written to tamper-evident ledger. Displayed to judges + athlete.
During pre-tournament calibration, the SEGA AI Judge scored 40 test performances alongside a panel of 3 UYSF-certified human judges. Results:
| Metric | Human panel | SEGA AI Judge |
|---|---|---|
| Mean absolute deviation across 40 performances | ±0.87 pts | ±0.24 pts |
| Inter-run consistency (same video, twice) | ±0.62 pts | ±0.00 pts (deterministic) |
| Viewing-angle sensitivity | up to ±1.4 pts (side vs front) | ±0.15 pts |
| Fatigue drift (hour 1 vs hour 6) | ±0.45 pts | ±0.00 pts |
| Correlation with medal outcome | 0.91 | 0.97 |
The AI is not designed to replace human judges — it standardises the measurement layer that human panels then interpret. Final tournament rulings remain human-led, with the AI providing a reference score of record.
Athlete or coach submits a written appeal at the scoring desk with the performance ID.
The original video is re-processed by the AI (deterministic — identical result) and reviewed by the head judge panel.
Head judge may adjust the score if landmark tracking clearly failed (e.g. costume occlusion). Override is logged with reason.
Adjusted score written to the tamper-evident ledger with the original + adjustment + reason. Full transparency.
Patent examination in Malaysia typically takes 2–4 years. The application publication and status may be verified through the MyIPO official register once published.
No. The SEGA AI Judge produces a standardised measurement score. UYSF-certified human judges are present at every station and retain override authority through the appeals process. The AI is designed to reduce inter-judge variance and viewing-angle differences, not to make the human panel redundant.
Yes — if the AI-measured Execution + Stability + Difficulty combination is identical to two decimal places. Tie-breaking follows the UYSF Rulebook (§ ranking clause): earlier submission wins the tie for ranking purposes.
The AI reports low-confidence landmarks. If confidence drops below the certification threshold on any critical joint, the scoring station flags the performance for human head-judge review before the score is committed.
The geometric ideals are proportional (measured relative to the athlete's own limb ratios), not absolute. A shorter or taller athlete performing the same asana with the same alignment quality will receive the same Execution score.
Certified scoring stations are available for licensing. Contact Prof Segar (KL) for the certification pathway. Stations must meet the camera + lighting + mat standards in §6.
An abstract was presented at AICSM 2025 at AIMST University, and a validation outline was shared with Universiti Malaya. See the Research page for the current publication list.
Open the full 2026 championship audit, or contact Prof Segar to arrange a demonstration for your school, federation, or research group.
Championship Audit → Contact Prof Segar