ADVANCING WUSHU TRAINING AND COMPETITION ANALYSIS: A DEEP LEARNING APPROACH TO AUTOMATIC ACTION DESCRIPTION IN MARTIAL ARTS

Abstract

Wushu is a kind of fitness activity loved by the broad masses of the people. It is one of the ways for people to cultivate their morality and cultivate their sentiment. In the process of martial arts competition, the action evaluation depends on manual work, and the results are susceptible to subjective factors such as the energy restriction of judges, and the objective and accurate completion quality score cannot be formed. At the same time, the Wushu movements are more esoteric. The automatic description of the competition movements in the Wushu competition can help the audience and the judges to understand and evaluate the Wushu movements. Therefore, it is necessary to introduce artificial intelligence technology to study a new martial arts action evaluation method, so as to provide accurate, intelligent and unlimited action evaluation for the majority of martial arts athletes. Based on this paper, a set of automatic description algorithm is proposed to provide accurate and robust evaluation results by extracting and modeling the features. This paper mainly selects the ornamental and representative Taijiquan in martial arts competitions to explain, mainly because the scope of martial arts is too wide. Firstly, the human gesture recognition technique of fusion deep transfer learning is used to obtain the key joint coordinates, according to the requirements of martial arts training outline, at last, construct the sample vector into the long-short memory neural network for training, the Taijiquan original library is constructed to ensure the implementation of the automatic description algorithm. In the experiment, the results show that compared with the traditional manual single-perspective evaluation method, this method can not only accurately identify the posture characteristics of the produced potential action, but also effectively quantify the strength of the action and objectively score the completion quality of the action, it can also further enhance the audience's experience when watching the video. This paper verifies the feasibility of combining artificial intelligence technology and martial arts competition, provides specific solutions for competitive sports competition in the era of artificial intelligence, and is an innovative practice in the field of artificial intelligence technology and sports competition. It can not only further improve the enjoyment of wushu competition, but also promote the promotion of martial arts sports, and make the decision of wushu competition more fair and objective.