Discovering and Translating Skills from Motion Data

To improve beginner athletes’ performance efficiently, they need to know how skilled athletes use their body parts. In our previous research, skills about a single body part were discovered, but we did not consider the relationship between skills of multiple body parts. Moreover, the feedback of skills to beginners was not presented. In this paper, we aim at discovering correlations between skills extracted from human motion data, and translate the skills to human understandable information. To extract skills from motion data, we compare the motions of skilled athletes and those of beginner athletes, and find the differences between them. For discovering skills about multiple body parts from motion data, we measure the correlation between body part movements using the “sliding window” method. By measuring and comparing the importance of the combinations extracted, we discover correlations between skills of multiple body parts. Translating skills is implemented by using Bayesian Networks that specify the states of motions and sentence templates that correspond to the states. Finally, we show that applying our algorithm to the motion data of the “overhead smash” in tennis can obtain results that are effective for improving beginner athletes’ performance.