Punch Type Classification and Hit Judgement Using Estimated Skeletal Model in Boxing Match Videos
Soma Watanabe, Yoshinari Kameda · 2024
In boxing, the choice of punch types and how to hit the punch to the opponent player is an important issue. So the support of computer vision on punch type classification and hit judgement on boxing match video is demanded. There are currently two challenges to that purpose. The first is the preparation of appropriate video dataset of boxing matches. The second is the discussion of the right method for punch type classification and punch hit judgment. We propose to create a video dataset of boxing matches from a boxing 3DCG simulation. The simulation can automatically annotate attributes to the dataset. This is useful for hit / no-hit judgement as it is not easy to identify which punches are actually hit and which are not on real boxing match videos. Based on the dataset we prepared, we propose a new method using time-series skeletal representation for classifying the type of punches and judging the hits. The experimental results show that our proposed method is able to classify thetypes of punches and judge the hits.