Pose Analysis and Classification in Shooting Sport Using Convolutional Neural Network and Long Short-Term Memory

Shooting sport requires high accuracy and speed, making training evaluation essential for athlete performance improvement. Conventional evaluation methods are often limited, thus the application of Artificial Intelligence (AI) and Computer Vision provides an effective alternative. This research aims to analyze and classify shooting sport poses using Deep Learning methods. A dataset consisting of several thousand pose images was collected from both field recordings and publicly available sources, followed by preprocessing for coordinate extraction. Convolutional Neural Network (CNN) was employed to extract coordinate data from shooting pose images, while Long Short-Term Memory (LSTM) was applied for pose classification. Experimental results demonstrated 94% accuracy, 95% Percentage of Correct Keypoints (PCK), and 4 mm Mean Per Joint Position Error (MPJPE), with training conducted at a learning rate of 0.0001 over 150 epochs on 5% test data, involving a total of 596,642 parameters. These results indicate that the proposed CNN–LSTM model provides a reliable approach for pose analysis and classification in shooting sport. The contribution of this study lies in presenting a novel dataset and framework for AI-based shooting sport evaluation, which can enhance training feedback and broaden AI applications in sports.