Publication:

Exploring the addition of audio input to wearable punch recognition

Martial arts can promote healthy lifestyles, improve self-confidence and provide self-defence skills. Previous work has demonstrated that inertial sensors can be used to recognise movements such as punches in boxing and support self-directed training. However, many martial arts do not use gloves which means that punches can be performed with different parts of the hand, and therefore produce a different sound on impact. We investigate if it is possible to recognise different punches executed with a bare hand, and if the recognition rate improves by combining audio input with the traditional inertial sensors. We conducted a pilot study collecting a total of 600 punches, using a wearable wristband to capture inertial data and a stand-alone microphone for audio input. The results showed that it was possible to distinguish five types of punches with 94.4% accuracy when using only inertial data, and that adding audio input did not improve the accuracy. These findings can guide the design of future wearables for punch recognition.