Research Lab

BlueX Research

Technical publications on intelligent systems for sports facility operations, edge-native computer vision, and semantic video understanding.

01
TIPS CertifiedJanuary 2026

Revenue Intelligence for Sports Facilities

TIPS-Certified Dynamic Pricing System

A machine learning system that treats sports facility reservations as perishable inventory, achieving 98% prediction accuracy through weather correlation, price elasticity optimization, and hybrid recommendation algorithms.

Accuracy: 98%
Modules: 3
Coverage: 16 Cities
02
ProductionJanuary 2026

BAVI Lycos X

Edge-Native Ball Trajectory Estimation

A lightweight neural architecture combining Temporal U-Net with ConvLSTM bottleneck for real-time ball detection. Achieves 75× parameter reduction over TrackNet while maintaining 98.2% recall at 5px tolerance.

Parameters: ~200K
Recall: 98.2%
Latency: 1.8ms
03
ProductionJanuary 2026

BAVI V-JEPA 2.0

Semantic Action Understanding

Latent-space video prediction architecture that eliminates pixel reconstruction overhead. Predicts masked video patches in embedding space, achieving 50% parameter reduction with 2.85× faster decoding than generative alternatives.

Parameters: 650M
Recall: 99.4%
Latency: 12ms
04
Case StudyJanuary 2026

Fast-Moving Object Detection

Iterative Self-Training for Sub-Pixel Tracking

Case study documenting our iterative training methodology for detecting fast-moving objects in sports video. Using tennis ball detection as primary domain, achieved >96% recall from 49.3% baseline through physics-based filtering and progressive dataset refinement.

Recall: >96%
Images: 42,863
Sources: 12
05
Case StudyJanuary 2026

Fighter Detection & Pose Analysis

Multi-Model Pipeline for Combat Sports

Multi-model pipeline combining custom YOLOv8 for fighter identification, YOLOv8-Pose for 33-landmark skeletal tracking, and ST-GCN for kick technique classification. Achieves real-time performance (<100ms latency) for taekwondo match analysis.

Latency: <100ms
FPS: 30+
Accuracy: 90%

Research Collaboration

For technical partnerships, model licensing, or academic collaboration inquiries.