Toward the Perfect Stroke: A Multimodal Approach for Table Tennis Stroke Evaluation

Abstract

Developing a consistent stroke is a challenge and even more so for nonprofessional Table tennis players. To build consistent proper strokes for beginner players, there is a need to understand the stroke differences between standard and beginner players. In terms of table tennis applications, prior works used a video-based method, or accelerometer sensor embedded in a table tennis racket, or infrared (IR) depth sensor for evaluating the stroke. However, there are certain challenges in these methods such as having insufficient data to analyse a complete stroke, time-consuming and costly data collection, as well as using non-prevalent equipment. Hence, to improve the beginner player's performance, an ubiquitous way through readily accessible commercial devices for stroke evaluation is essential. In this study, to achieve such a goal, we (i) recorded videos and accelerator signals of standard and beginner players using consumer-grade products, and (ii) analysed the stroke consistency of both standard and beginner players. The results show the significant differences in the strokes between both kinds of players through the multimodal approach. Also, we found the significantly strong correlation between the stroke consistency and the hitting score for the forehand stroke. These findings motivate us to further examine the improvement of beginner players by instructing procedural knowledge of a standard player's stroke, and implement applications for the motor-skill instruction.

A Survey on Video Action Recognition in Sports: Datasets, Methods and Applications

“…On one hand, recognizing the group activities is able to assist coaches to make better decisions and players to understand their performances on fulfilling the coaches' strategies. On the other hand, recognizing the individual actions can benefit training players via correcting the small action errors [31], [32]. Another wide application of sports action recognition is in sports TV programs, where there is a massive demand of highlights generation and action recognition can significantly improves the localization accuracy [33]- [36].…”

$\textbf{P$^2$A}$: A Dataset and Benchmark for Dense Action Detection from Table Tennis Match Broadcasting Videos

“…There are many applications of video understanding, such as recommendation systems [7,30,63]. One of the most attractive applications is understanding sports videos, which is able to benefit coaching [3,52], player training [4,46] and sports broadcasting [28,53,53]. To better understanding sports videos, localizing and recognizing the actions in untrimmed videos play crucial roles, however, they challenging tasks with unsolved problems.…”