An Analysis Method of Tactics in Karate Matches Using a Bayesian Network Model

This study proposes a model using a Bayesian network to understand tactical behavior in Karate matches. The model is a probabilistic causal model consisting of the states of two competitors engaged in combat. Each state node of the model outputs a probability distribution of the occurrence of offensive, defensive, and evaluative actions. Using the model, we also propose an analysis method of Karate tactics that can obtain the high likelihood processes for a competitor for the case in which the competitor’s offensive action ends in success. The extracted processes indicate competitor’s tactical pattern. For an experiment of the method, we collected action data from the videos of Karate matches in which elite competitors had competed, and trained two elite competitor’s model. The method has extracted the elite competitors’ tactical actions correctly using the trained models.