Fig 1.
Schematic diagram of the overall process of AI table tennis coaching.
Fig 2.
Outlines the workflow of table tennis trajectory recognition.
A:original video. B:background subtraction and Gaussian blur. C:ball extraction. D:ball motion trajectory localization and recognition.
Fig 3.
The process of motion capture and analysis for table tennis.
A:OpenPose motion capture of a table tennis player during a stroke. B:motion charts derived from the capture, highlighting changes in shoulder and arm speeds, shifts in body center coordinates, arm angles, shoulder positioning, and body angle variations.
Table 1.
Corresponding multimodal input data for each error type.
Fig 4.
A: Schematic of a Large Language Model Based on a Vector Knowledge Base. B: Example of prompts for identifying arm related errors.
Table 2.
Corresponding multimodal input data for each error type.
Fig 5.
A:The accuracy and false positive rates of the AI table tennis coaching system in judging various types of errors in table tennis. B: Confusion matrix of AI table tennis coaching system for various errors.
Table 3.
Comparison of AI table tennis coaching system and sole use of GPT for error detection.
Fig 6.
AI table tennis coach analysis and guidance on a novice’s point loss process.
Table 4.
Details of expert evaluation of the AI table tennis coaching system.
Fig 7.
Automated loss analysis, error cause analysis and recommendations for a table tennis novice in an 11-point match.
Fig 8.
Error recognition performance of different multimodal large language models.
Fig 9.
Average token usage and costs in various processes of AI table tennis coaching system.