Optimization Study of Traditional Martial Arts Movement Recognition Algorithm Incorporating OpenPose

In the domain of traditional martial arts motion recognition, the precise capture and analysis of human posture have become pivotal to enhancing recognition accuracy. In recent years, with the rapid advancement of deep learning technology, algorithms for action recognition based on skeletal keypoints have gradually become a focal point of research. However, existing methods exhibit suboptimal performance under complex postures and occlusion scenarios, particularly in the context of traditional martial arts, where the actions are varied and rapidly changing. As a high-performance human pose estimation algorithm, OpenPose addresses this challenge by accurately pinpointing human keypoints and generating high-quality confidence maps. This paper delves into the network architecture of OpenPose and its methodologies for human region detection and skeletal keypoint association, aiming to optimize and integrate OpenPose into an action recognition system to improve the accuracy and robustness in traditional martial arts motion recognition. Through a comparative analysis of mainstream neural network approaches such as AlexNet, GoogLeNet, and ResNet, along with an exploration of skeletal keypoint-oriented action recognition algorithms like AGC-LSTM and DPRL, this paper proposes an action recognition system framework based on OpenPose. Extensive experimental simulations not only validate the feasibility of this system but also further examine its performance across various environments. The ultimate goal of this research is to provide an efficient and reliable solution for traditional martial arts motion recognition, thereby promoting the inheritance and development of martial arts culture.