Learning Three Dimensional Tennis Shots Using Graph Convolutional Networks

Human movement analysis is very often applied to sport, which has seen great achievements in assessing an athlete’s progress, giving further training tips and in movement recognition. In tennis, there are two basic shots: forehand and backhand, which are performed during all matches and training ses...

Full description

Bibliographic Details
Main Authors: Maria Skublewska-Paszkowska, Pawel Powroznik, Edyta Lukasik
Format: Article
Language:English
Published: MDPI AG 2020-10-01
Series:Sensors
Subjects:
Online Access:https://www.mdpi.com/1424-8220/20/21/6094
Description
Summary:Human movement analysis is very often applied to sport, which has seen great achievements in assessing an athlete’s progress, giving further training tips and in movement recognition. In tennis, there are two basic shots: forehand and backhand, which are performed during all matches and training sessions. Recognition of these movements is important in the quantitative analysis of a tennis game. In this paper, the authors propose using Spatial-Temporal Graph Neural Networks (ST-GCN) to challenge the above task. Recognition of the shots is performed on the basis of images obtained from 3D tennis movements (forehands and backhands) recorded by the Vicon motion capture system (Oxford Metrics Ltd, Oxford, UK), where both the player and the racket were recorded. Two methods of putting data into the ST-GCN network were compared: with and without fuzzying of data. The obtained results confirm that the use of fuzzy input graphs for ST-GCNs is a better tool for recognition of forehand and backhand tennis shots relative to graphs without fuzzy input.
ISSN:1424-8220