With the vigorous development of computer vision and sensor technology, the acquisition and storage of video data become easier and easier, which promotes the development of human body recognition technology. Chinese martial arts are profound and profound, with the Chinese civilization inherited until now, the application of modern technology to promote and inherit traditional martial arts has attracted much attention. In this paper, we study a martial arts action recognition system based on time series analysis, feature extraction, feature representation and action recognition of martial arts actions. By comparing the action recognition rates of standard action demonstrators and testers, we prove that the recognition system can effectively describe and distinguish different martial arts actions, which is of reference significance for the enhancement of the accuracy of martial arts actions.