SEGA AI Judge System · Patent Pending PI2026005219 · Deployed at 2026 Championship
Home  /  AI System  /  Evidence Centre
// SEGA AI Judge™ · Evidence Centre

The full evidence base behind the world-first AI-judged Yogasana championship

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.

Malaysian Patent Application
PI2026005219
Status: Pending · Filed 2026 · Inventor: Prof Segar

Jump to a section

§1 · The invention

Why an AI Judge for Yogasana was necessary

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.

§2 · Technology stack

What sits inside the system

Layer 1

Video Capture

1080p 30fps front-facing camera + optional 45° side camera. Two-camera synchronisation via NTP timestamp.

Layer 2

Skeleton Tracking

Real-time human pose estimation extracts 33 body landmarks per frame (head, shoulders, elbows, wrists, hips, knees, ankles, feet, hands, fingertips).

Layer 3

Landmark Detection

Sub-pixel refinement on critical joints — spinal midline, sacral tilt, cervical curve — using SEGA's proprietary Yogasana-specific model.

Layer 4

Geometric Comparison

Each landmark set compared against the ideal geometry for the declared asana, indexed by the UYSF Tariff Table.

Layer 5

Temporal Stability

Frame-to-frame variance of landmark positions during the hold window, quantified as CoG micro-tremor.

Layer 6

Scoring Aggregator

Three sub-scores (Execution, Stability, Difficulty) combined with the tariff multiplier to produce the final 0.00–10.00 score.

§3 · Scoring pipeline

From video-in to score-out in 4 seconds

Second 1

Ingest

Athlete video streamed to the scoring station. Frame extraction at 30fps.

Second 2

Pose extract

Skeleton tracker outputs 33 landmarks × N frames.

Second 3

Compute

Geometric & temporal analysis. Execution + Stability + Difficulty sub-scores.

Second 4

Aggregate + log

Final score written to tamper-evident ledger. Displayed to judges + athlete.

§4 · Scoring formula

The exact mathematics

50%
ExecutionDeviation of every landmark from ideal geometry for the declared asana.
E = 10 − Σ|Δ_i|·w_i
30%
StabilityFrame-to-frame landmark variance during the 30s hold window.
S = 10 − k·σ_hold
20%
DifficultyAsana tariff (0.5–2.0×) from the UYSF pose-tariff table.
D = tariff × 10

Final score composition

Weighted meanFinal = (0.5·E + 0.3·S + 0.2·D)
Range0.00 to 10.00 (two decimals)
Tremor penalty−0.25 per detected wobble event beyond baseline (per UYSF Rulebook clause 4.2)
Difficulty capA perfectly executed simple posture cannot outscore an equally well-executed advanced posture (audit §16 protection)
§5 · Validation

SEGA AI vs. human judge panel

During pre-tournament calibration, the SEGA AI Judge scored 40 test performances alongside a panel of 3 UYSF-certified human judges. Results:

MetricHuman panelSEGA 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 sensitivityup 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 outcome0.910.97

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.

§6 · Camera requirements

What venues need

Minimum kit for a certified scoring station

  • Front-facing camera: 1080p @ 30fps minimum, fixed tripod, 3m from athlete mat
  • Optional 45° side camera for spinal alignment verification (recommended for International grade)
  • NTP-synchronised time server (accurate to ±10ms) for two-camera fusion
  • Ethernet or 5GHz Wi-Fi to scoring station (avoid 2.4GHz — competition venues have interference)
  • Backup local recording on SD card at each camera (in case network fails, video is preserved)
  • Uniform mat colour (SEGA-standard olive green or similar) for consistent skeleton contrast
  • Lighting: 500–1000 lux, diffuse overhead, no direct backlight (window behind athlete disqualifies footage)
§7 · Championship audit

2026 event evidence record

Kandiah Hall · Kuala Lumpur · 18–19 July 2026

Event statusCompleted (EventCompleted per Schema.org)
Participants128 athletes
Nations represented9 (🇲🇾 🇮🇳 🇸🇬 🇹🇯 🇱🇰 🇮🇩 🇹🇭 🇳🇵 🇦🇪)
Medals awarded113 (gold + silver + bronze across all categories)
AI-judged performancesEvery performance was AI-scored in real time
Human oversightUYSF-certified judge panel present at every station
Independent witnessesIndian High Commission (Mr Prateesh Kumar, Counsellor Community Affairs); UYSF World President (Shri Suresh Kumar)
DocumentationFull audit report + rankings preserved
View full Tournament Audit → View live Rankings → Peer-reviewed Research →
§8 · Appeals process

What happens if an athlete disputes a score

Within 15 min

File appeal

Athlete or coach submits a written appeal at the scoring desk with the performance ID.

Review

Video re-run

The original video is re-processed by the AI (deterministic — identical result) and reviewed by the head judge panel.

Human override

Panel decision

Head judge may adjust the score if landmark tracking clearly failed (e.g. costume occlusion). Override is logged with reason.

Final

Ledger update

Adjusted score written to the tamper-evident ledger with the original + adjustment + reason. Full transparency.

§9 · Data security

How performance data is protected

Chain-of-custody controls

  • Every performance video hashed (SHA-256) at capture; hash written to ledger before scoring begins
  • Score record + video hash + timestamp signed by scoring station on write; no post-hoc edits possible without breaking the signature chain
  • Athletes under 18 have full biometric protection: video retained for 90 days only (appeal window), then deleted
  • Row-Level Security on all athlete records; only the athlete and SEGA administrators can read personal fields
  • Public rankings show display name + country + score only; DOB, contact, home address never public
  • Adult athletes may opt out of public profile display — scores still recorded but not shown by name
§10 · Patent status

Malaysian patent application PI2026005219

Patent record

Application numberPI2026005219
Filing jurisdictionMalaysia (MyIPO)
Filing year2026
StatusPending examination
InventorProf Segar (SEGA Intercontinental Yoga Alliance)
Claim scopeMethod and system for computer-vision-based competitive Yogasana scoring, comprising landmark-based execution measurement, temporal stability quantification, and tariff-weighted difficulty aggregation.
Prior art distinctionUnlike general pose-estimation systems (MediaPipe, OpenPose), the SEGA method is domain-specific to Yogasana geometry with proprietary hold-window stability quantification and UYSF-tariff integration.

Patent examination in Malaysia typically takes 2–4 years. The application publication and status may be verified through the MyIPO official register once published.

§11 · Frequently Asked Questions

Common questions from athletes, coaches, federations

Does the AI replace human judges entirely?

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.

Can two athletes get identical scores?

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.

What if the camera misses a landmark (e.g. long sari occluding an ankle)?

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.

Is the AI biased against certain body types?

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.

How can a school run its own SEGA AI-judged tournament?

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.

Where is the peer-reviewed research?

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.

Want to see the AI Judge in action?

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