Projects
Computer Vision
EXPERIMENTPipeline validation · updated 2026-07-19Muay Thai Data & AI Platform
แพลตฟอร์มข้อมูลและเอไอมวยไทย
Fight analytics from video: fighter identification, pose estimation, strike detection, clinch analysis and multi-camera statistics for Muay Thai.
ระบบ AI วิเคราะห์การแข่งขันมวยไทยจากวิดีโอหลายมุมกล้อง
View ResearchDemoArchitecturePaperDatasetCodeAPI
Problem
Muay Thai analysis is manual and inconsistent; occlusion in clinch and rapid strike sequences defeat generic action recognition models.
Research Question
Can pose-conditioned action recognition reliably separate strike classes during clinch occlusion?
Objectives
- Identify red and blue corner fighters robustly across rounds
- Detect and classify strikes with temporal grounding
- Produce per-round fight statistics from multi-camera input
Methodology
- 01Camera
- 02Video Processing
- 03Detection
- 04Segmentation
- 05Tracking
- 06Pose Estimation
- 07Action Recognition
- 08Fight Analytics
- 09Dataset
Architecture
Multi-camera InputRTSP / file
Synchronised streams
Detection & TrackingYOLO + tracker
Persistent fighter IDs
Pose EstimationMediaPipe
Skeleton per fighter
Action RecognitionTemporal model
Strike and clinch classes
AnalyticsPostgreSQL
Round-level statistics
Results & Benchmark
- Tracked entities
- 2 fighters + referee
Limitations
- · Clinch occlusion remains the dominant error source