Quantifying hitting activity in tennis with racket sensors: new dawn or false dawn?

The non-invasive measurement of movement through sensor technologies is common in elite sport, yet only recently become possible in tennis. This study validated two commercial tennis racquet sensors compared to gold standards in VICON and expert notational analysis. One national-level male tennis player hit 24 shots (9 forehands and 9 backhands, hit with heavy topspin, flat and slice; 6 serves) using a Babolat Play racket with a Zepp sensor. Measures of shot type, impact location and racket speed were compared against those captured by a 500-Hz 12 camera VICON system. Cohen's kappa (κ) and a weighted kappa (κw) assessed the accuracy of stroke classification and impact location, respectively, and intraclass correlation coefficients (ICC) validated sensor-derived racket speed. Both sensors recorded the same total stroke volume as VICON, while Zepp racket speed displayed near perfect agreement with VICON (ICC = 0.983; p < 0.001). They exhibited moderate agreement with true stroke classification (Babolat: κ = 0.730; Zepp: κ = 0.612) and minimal agreement with true impact location (Babolat: κw = 0.412; Zepp: κw = 0.217). During match play, both sensors achieved near perfect accuracy for stroke volume but experienced difficulty discriminating the different stroke types. In sum, the Babolat Play and Zepp determined stroke volume and intensity accurately but were less effective in identifying specific strokes and impact locations.