Pose-Based Binary Shooting Posture Classification Using YOLO-Pose and SVM

Human pose estimation has been widely applied in various fields such as sports performance assessment and health data analysis. Firearm training can also benefit from this technology, as correct shooting postures improve both accuracy and trainee safety. This study presents a pose-based approach for binary classification of standing shooting postures, determining whether a posture is correct or incorrect. The proposed method employs a deep learning-based Human Pose Estimation (HPE) model that extracts key body points (keypoints) and their confidence scores using YOLO-pose models. These extracted features are combined into a feature vector and classified using a Support Vector Machine (SVM) with different kernel functions, namely: Linear, Polynomial, RBF, and Sigmoid. Experimental results show that the Linear and RBF SVMs combined with YOLOv8n-pose or YOLOv11n-pose models achieve accuracies and F1-scores above 80%. The highest performance, reaching 87.34% Accuracy and 87.27% F1-score, is achieved by the Linear SVM using YOLOv11n-pose features extracted from rear-view camera images. Therefore, the combination of practical YOLOpose feature extraction with the highly interpretable Linear SVM establishes the proposed method as a promising, efficient, and interpretable real-time solution for shooting posture evaluation.