Table tennis motion recognition based on the bat trajectory using varying-length-input convolution neural networks

Abstract

Action recognition has been applied in fields such as smart homes, gaming, traffic management, and security monitoring. Motion recognition is helpful for biomechanical analysis, auxiliary training systems, table tennis robots, motion-sensing games, virtual reality and other fields. In our study, we collected data on table tennis skill motion, created the TTMD6 dataset, and analyzed the characteristics of table tennis paddle trajectories. We propose a motion recognition algorithm to recognize paddle trajectories. Other research has used multijoint data to identify actions, while we use only the paddle trajectory to recognize table tennis skill motions, accelerating the speed of motion recognition. Therefore, it is feasible to use paddle trajectories to recognize table tennis skill motions.