Automatic Evaluation of Boxing Techniques from Captured Shadow Boxing Data

We propose a fully automatic procedure to preprocess captured shadow boxing motion and evaluate the boxer's techniques. The captured shadow boxing motions are automatically segmented and classified into different actions such as straight punches, hooks, or translations. Then, it is possible to automatically evaluate their skills by checking the quality of their movements and finding out which kind of motions are available by the boxers. It is also possible to compose a Motion Graph of the fighters and simulate the matches of different boxers to find out their weak points. The proposed method is useful for boxers to periodically check their performances.