SYSTEM AND METHOD OF SPORTS TRAINING USING MACHINE LEARNING AND GENERATIVE AI

Smart overview of the Invention

A novel system for enhancing athletic performance through AI leverages machine learning and generative AI technologies. By analyzing player videos, the system extracts skeletal and temporal data, aligning it with human commentary. This information is processed using transformer-based or large language models (LLMs) to understand the relationship between motion patterns and coaching semantics. The system offers textual feedback and visual reconstructions to illustrate improved techniques, supporting personalized training across various sports.

Technical Field and Related Technology

The system addresses challenges in sports training, where techniques are not simply right or wrong but exist on a spectrum influenced by individual physique. Traditional methods involve video overlays and physical markers, which can be limited by the complexity of biomechanics. This AI-driven approach offers a more nuanced understanding, accommodating diverse body types and techniques, enhancing the effectiveness of coaching methods.

Generative AI in Sports

The integration of generative AI, large vision models (LVMs), and multimodal models into sports training represents a significant advancement. This approach is applicable across various sports, from racket sports to gymnastics, providing insights into movement-based techniques. AI models undergo training and inference phases, using player performance data and coaching inputs to refine and apply techniques, offering comprehensive coaching outputs.

System Architecture

The system's architecture includes several modules: a player video database, pre-processing, data annotation and key feature extraction, input data selection, and AI model training. Videos are standardized, annotated, and transformed into structured data for AI training. This enables the model to learn from diverse datasets, improving its ability to provide actionable feedback and technique corrections.

Data Processing and Model Training

Data processing involves standardizing video inputs, extracting key features using computer vision algorithms, and organizing data for model training. Annotated data is structured into feature matrices, capturing motion details essential for AI learning. The system ensures a balanced dataset, integrating new data while maintaining consistency, facilitating robust model training and effective sports training applications.