The utility of markerless motion capture for performance analysis in racket sports

Abstract

Recent technological advancements have allowed movements to be tracked ecologically via markerless motion capture (mocap). However, occlusions remain a major concern pertaining to markerless mocap. Within racket sports where the number of players involved are low and occlusions are minimal, there exists a unique opportunity to delve into and provide an overview on the utilisation of markerless mocap technology. Twenty studies were included after a systematic search. Several methods were applied to obtain 2D positional data. Most studies adopted some form of background subtraction or thresholding method ( n = 12), the remaining relied on pose estimation algorithms (PEA; n = 3), Hawk-Eye ( n = 2) and object recognition ( n = 1). Conversely, only the visual hull method was found to obtain 3D joint kinematics ( n = 2). Markerless mocap are conventionally used to extract joint kinematics, however, study results revealed that the predominant use of markerless mocap was to capture the movement of a player’s location on court, this finding was unexpected. Low sampling frequencies of input videos and unsuitability of model detection used in the included studies could have limited the ability for markerless mocap to accurately track movements in racket sports. While current evidence suggests that the use of PEA in racket sports to extract 3D kinematics is limited, perhaps a slightly different approach gearing towards performance analysis, specifically stroke classification with the amalgamation of player location data and joint kinematics may be worth exploring further.