A sport recognition method with utilizing less motion sensors

Abstract

In this study, we propose a recognition method in ball games using no more than two triaxial accelerometers on the user’s front arm and upper arm to track motion data. To produce effective features for classifying ball games’ postures, the motion data is processed by our method, which includes a median filter, a duplication removal algorithm, and an algorithm of feature extraction. Subsequently, the produced features are recognized by a support vector machine scheme for sports with single-handed swings like tennis, badminton, and ping pong. The research result in this investigation can help the athlete training of the above mentioned sports. Experimental results showed that the precision rate of the proposed method for recognizing postures in a single-handed swing achieves 95.67%.