Answers are grounded in this paper and cite their sources.
Action recognition in sports is essential in improving the fan experience and analysis of athletic performance, especially in dynamic environments like hockey. Current methods often can’t effectively capture detailed motion since they lack feature fusion from vision and bone data. The research seeks to create a sophisticated framework for hockey action recognition and classification by combining deep models and feature fusion methods, enhancing the recognition process and immediate usability in sports analytics. The developed system trains a deep learning model on the Hockey Action Recognition Pose Estimation dataset by integrating CNN, GCN, and feature fusion techniques. CNN extracts spatial features, and GCN extracts spatial interactions between joints to learn the complex human body motion. Feature fusion is used for fusing the output of the two models to gain better action classification performance. The model achieved very high recognition rates for the most critical hockey actions. The proposed system achieved an overall accuracy of 98.73% in recognizing the hockey actions “Forward,” “Backward,” “Passing,” and “Shooting,” demonstrating its effectiveness in detecting complex movements and providing reliable action recognition performance. The model achieved very high recognition rates for the most critical hockey actions. The model uses Python as the programming language, employing packages such as TensorFlow for deep learning processes. This design significantly enhances the detection of sports action towards reaching dynamic analysis for sport analytics, performance assessment, and tactical decision-making support. Its usage can significantly enhance professional coaching and live-event telecasting.
No takes yet. Share an insight, caveat, or question.
Peng et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: