Analysis of Human Body Posture During Ice Sports Based on Deep Learning

Abstract

To better analyze human posture in ice sports, a method based on deep learning has been established. In this method, the input gate, forget gate and output gate are configured to make them more suitable for human pose analysis relying on the LSTM network framework. The input data of LSTM are encoded using the sequence of human skeleton and joint angles in ice skating. During the experiment, abstract models of human bones and joints were provided, with continuous jumping movements in figure skating as the analysis object. LSTM provided detailed analysis results. The experimental results show that the accuracy of deep learning continues to improve in analyzing human posture during ice sports with the number of iterations increasing.