Design of volleyball action recognition method integrating human posture and spatiotemporal graph convolution

With the rise of intelligent algorithms, automated motion recognition in sports like volleyball—known for its complex movements—remains challenging. To enhance accuracy, this study introduces the Human Posture and Spatiotemporal Graph Convolution (HP-SGC) model. It uses skeletal keypoints to create a 2D coordinate system, combines object detection and pose estimation for action recognition, and applies spatiotemporal graph convolution for classification. The tests showed over 95% accuracy for continuous actions, with recognition time as low as 2.2 seconds. The model successfully identified 8 foul actions, 7 basic moves, and 4 skillful actions, averaging >90% accuracy. These results demonstrate HP-SGC’s strong performance in volleyball action recognition, offering valuable tools for match analysis and statistics.