Taekwondo Action Recognition Method Based on Partial Perception Structure Graph Convolution Framework

Abstract

Action recognition in Taekwondo competitions and training is an important task, which can provide a very valuable reference factor for technicians, athletes, and coaches. We propose a graph convolution framework with part of the perception structure to recognize, decompose, and analyze Taekwondo actions. Taking advantage of the long short-term memory of a part of the perception structure, the recognized Taekwondo actions are marked in time series, and then features are extracted from the graph convolution level to obtain the spatial and temporal associations between joints. Predict the action category and perform score matching based on the manual tag database. Finally, it is verified on our self-made Taekwondo competition data set. Our method has an average accuracy of 90% in action recognition, and an average action score matching rate of 74.6%. The accuracy of action recognition is high, which provides great assistance to Taekwondo e training and competitions.