Volleyball Pose Recognition System Based on DLSTM-GCN Algorithm

Abstract

Aiming at the current problems of poor volleyball pose recognition and low recognition accuracy, this study proposes a volleyball pose recognition system based on dual long short term memory network graph convolutional network. A volleyball pose recognition system based on dual long short term memory network graph convolutional network is proposed. The new system firstly estimates and analyzes the volleyball sports data by multi-scale feature extraction and fusion network. Secondly, dual long short term memory network graph convolutional network is introduced to analyze the estimated video data. The outcomes revealed that among the different algorithmic model losses, multi-scale feature extraction and fusion network had the smallest loss value, which was reduced by 0.011 compared to the OpenPose algorithm. Meanwhile, the improved dual long short term memory network graph convolutional network was able to achieve the highest recognition accuracy of 93.68%, which was improved by 2.00% compared to the unimproved one. The running time of the improved dual long short term memory network graph convolutional network was shorter at only 2.1s, and the recall was also able to reach up to 92.68%. Meanwhile, in the actual running test of different algorithms, the improved dual long short term memory network graph convolutional network had more key point detection and more specific image action. In summary, the improved algorithm has better volleyball action recognition effect, which has a better application effect on volleyball sports pose recognition and volleyball action guidance research.