A fencing action recognition algorithm based on keypoint detection
Qi Jin, Yong Tian Zhu · 2025
In recent years, human motion recognition technology has made remarkable progress in the field of artificial intelligence vision, especially in the evaluation of athlete performance and training optimization. As a sport that emphasizes basic technical movements and steps, foil's traditional teaching mode relies on the subjective judgment of coaches. However, due to the limited number and uneven level of coaches, beginners often face difficulties in movement correction, slow progress and injury risks. Therefore, combined with the foil movement recognition technology in artificial intelligence, it can provide students with scientific personalized guidance. Based on deep learning and combining OpenPose human pose estimation algorithm with convolutional neural network (CNN) and short term memory network (LSTM), this paper designs a combination model of OpenPose and CNN-LSTM. By extracting the key points of the human body and converting them into digital feature sequences, the model effectively reduces the input data dimension, eliminates the background noise interference, optimizes the network structure, improves the model training efficiency, and the recognition accuracy rate reaches 96.29%.