Exploration of Gymnastic Movement Analysis and Skill Improvement Paths Based on Long and Short Term Memory Networks

Abstract

For a long time, the diversity and complexity of gymnastics movements have made it more difficult to carry out the analysis of gymnastics movement evaluation, which is an obstacle to the improvement of gymnastics skills. In order to improve the recognition effect of gymnastics movements, this paper combines OpenPose and LSTM to establish a HOPL model for gymnastics movement recognition and evaluation. The model combines the original OpenPose network with a high-resolution network to achieve multi-feature fusion of gymnastic actions, and introduces the LSTM model to recognize or predict complex actions and behaviors by using the temporal information of skeleton sequences. Experiments show that the recognition accuracy and global F1 score of the HOPL model are 90.75±1.76% and 86.72±0.93%, respectively, and the computational consumption is low, so it is feasible to apply it to the evaluation of gymnastic movements. Relying on the application of deep learning technology in the field of gymnastics movement recognition, it can provide optimization strategy support for gymnastics movement skill improvement.