Commonality of Motions Having Effective Features with Respect to Methods for Identifying the Moves Made during Kumite Sparring in Karate.

In recent years, active use of imaging technology, sensor technology, and AI (artificial intelligence) has been on the rise to identify various plays that occur in sports competition to assist judges, coaches, and athletes. However, no study result has been reported on this research topic as it pertains to karate sparring competition. The lack of past studies on this topic is attributable to the fact that no viable method has been developed to acquire motion data or to identify athletes' motions at a fundamental level in competition, as no sensors can be worn by athletes on their bodies, and there are a number of blind spots that occur in competition due to fast-paced exchanges that competing athletes engage in, among other factors. Therefore, in this study, footage of kumite (sparring) in the practice of karate simulating actual competitive matches was captured using video cameras to conduct a motion identification experiment using a CNN (convolutional neural network). After comparing the result of this study to that of previous motion identification experiments in which subjects wore sensors on their bodies, it has been determined that the motions having effective features are common between the two types of experiments.