Automatic Activity Classification Based on Human Body Kinematics and Dynamic Time Wrapping

Abstract

Human movement analysis often relies on obtaining and processing digital signals from the lab-based biomechanical equipment such as motion capture system and force plate. This paper introduced a machine-learning based method, known as the Dynamics Time Wrapping (DTW) for human movement analysis. The DTW is used to classify four basketball playing movements including shoot, layup, dribble and pass. The kinematic raw data were obtained during an experiment session. The sample kinematic data were selected and normalized to create the templates. The DTW compared the kinematic data from each movement with the template. A 3-fold cross validation was used to validate the method. The results show that this method can achieve a high activity classification accuracy.