High Recognition Accuracy Achieved in SVM-Based Action Recognition Model
Researchers from Chengdu Sport University in China have developed an SVM-based action recognition model that utilizes skeletal key point analysis with posture sensors to provide accurate sports training analysis tools. The study, published in Bmc Sports Science Medicine and Rehabilitation, employed the quaternion method to model human skeleton features, acquired motion data through a posture sensor, and performed preliminary data processing using the Kalman filtering technique. The model achieved a high recognition accuracy of 91.24% for both static and dynamic actions.
Key Takeaways:
- The study proposes an SVM-based action recognition model that integrates human action recognition in sports training, aiming to provide an accurate sports training analysis tool.
- The research model utilizes skeletal key point analysis with posture sensors, which acquires motion data and performs preliminary data processing using the Kalman filtering technique.
- The model achieves a recognition accuracy of over 90% for static actions and over 80% for dynamic actions, with an average recognition accuracy of 91.24%.
- The high recognition accuracy of the model provides effective technical support for action improvement and technical analysis in sports training.
- The study demonstrates the reliability and validity of the human action recognition model proposed.
- The model can be used to help sports training by effectively distinguishing the feature points of different actions.
Statistics:
- Recognition accuracy for static actions: 90.12%
- Recognition accuracy for dynamic actions: 81.42%
- Average recognition accuracy: 91.24%
- Number of experimental subjects: 50
- Number of actions recognized: 20
- Time period of data acquisition: 1 hour
Sources:
- Bmc Sports Science Medicine and Rehabilitation, 2025;17(1):220
- Yixuan Cao, Chengdu Sport University, Chengdu, 610299, People's Republic of China
- Bmc, Campus, 4 Crinan St, London N1 9XW, England