Efficient characterization of tennis shots and game analysis using wearable sensors data

Recent trends show that wearable devices with high-range inertial sensors are actively being used for outdoor activities. The paper describes our developed sports analytics engine used for self-learning and/or coach-assisted training for swing-based games like tennis, golf, etc. by utilizing rich set of data collected from these wearable sensors. The sports analytics engine for tennis uses techniques based on modified Pan-Tompkins algorithm for detecting the shot and then uses time-warping based hierarchical shot classifier which uses Dynamic Time Warping (DTW) at first level (forehand, backhand and serve) and Quaternion Dynamic Time Warping (QDTW) at second level (slice and non-slice). Major challenges included distinguishing shots from noise in sensor data, classifying the shots based on information only from wrist of player and capturing the various playing styles across different players. Based on efficacy of the developed engine, we foresee wider usages of the proposed techniques in developing learning applications for swing-based sports.