Boxing Strike Classification from Wearable IMU Sensor Unit

This paper discusses the development of a boxing trainer system and its test results. The accelerometer and gyroscope were utilized in training to classify two different strikes over three activities by wearing an IMU in a boxing glove. A Metaware MetaMotion R device was used to collect the data. The results from the test were verified with a machine learning algorithm to verify the accuracy of the system. The system used KNN to classify the two different punches. The test shows that the model showed $99.9 \%$ accuracy in training and $73 \%$ accuracy in testing. When successive strikes of the same strike are performed, the system is able to classify with high accuracy, but in a real scenario testing, a combination of jabs and hook punches resulted in a lower classification performance.