Validating an algorithm from a trunk-mounted wearable sensor for detecting stroke events in tennis

ABSTRACT

This study analysed the accuracy of a prototype algorithm for tennis stroke detection from wearable technology. Strokes from junior-elite tennis players over 10 matches were analysed. Players wore a GPS unit containing an accelerometer, gyroscope and magnetometer. Manufacturer-developed algorithms determined stoke type and count (forehands, backhands, serves and other). Matches were video recorded to manually code ball contacts and shadow swing events for forehands, backhands and serves and further by stroke classifications (i.e., drive, volley, slice, end-range). Comparisons between algorithm and coding were analysed via ANOVA and Bland-Altman plots at the match-level and error rates for specific stroke-types. No significant differences existed for stroke count between the algorithm and manual coding (p > 0.05). Significant (p < 0.0001) overestimation of “Other” strokes were observed from the algorithm, with no difference in groundstrokes and serves (p > 0.05). Serves had the highest accuracy of all stroke types (≥98%). Forehand and backhand “drives” were the most accurate (>86%), with volleys mostly undetected (58–60%) and slices and end-range strokes likely misclassified (49–51%). The prototype algorithm accurately quantifies serves and forehand and backhand “drives” and serves. However, underestimations of shadow swings and overestimations of “other” strokes suggests strokes with reduced trunk rotation have poorer detection accuracy.

KEYWORDS:

• Racquet sports
• accuracy
• external load
• wearable sport technology

Disclosure statement

The authors would like to declare that Thomas Perri, Machar Reid and Alistair Murphy are employed by Tennis Australia and Kieran Howle is employed by Catapult Sports.

Additional information

Funding

The author(s) reported there is no funding associated with the work featured in this article.