Developing Intelligent Sports Analysis System: Enhancing Keypoint Prediction, Object Tracking, and Computational Efficiency in Real-World Scenarios
- Publication Type:
- Thesis
- Issue Date:
- 2025
Open Access
Copyright Clearance Process
- Recently Added
- In Progress
- Open Access
This item is open access.
This thesis introduces an intelligent sports analysis system that addresses key challenges in objective sports examinations. To tackle inaccurate human keypoint predictions caused by occlusions, the system employs Spatio-Temporal Graph Neural Processes (STGNP), which use cross-set graph neural networks, causal convolutions, and Graph Bayesian Aggregation (GBA) to infer missing skeleton data accurately. For dynamic examination scenarios like basketball skill assessments, an enhanced object tracking model is developed. By integrating additional candidate positional information, the system improves tracking robustness, even when equipment such as basketballs and cones become occluded. Furthermore, the thesis proposes a computationally efficient approach for skeleton-based motion recognition. This method leverages Lie algebra for representing skeletal data, and combines LSTM with CNNs, embedding the network weights into a memristor-based structure for faster inference and reduced energy consumption. Extensive experiments validate that these innovations significantly enhance accuracy, reliability, and efficiency in real-world sports examination environments.
Please use this identifier to cite or link to this item: