A computational framework for athlete training assistance system based on health big data analysis

Precise improvement of athlete training effectiveness relies on the synergistic optimization of motion technique analysis and health status monitoring. To address issues such as insufficient motion recognition accuracy and delayed integration of health data in key technical training (e.g., tennis serving), this study constructs a computational framework for training assistance, integrating computer vision and health big data analysis. The framework consists of three core computational modules: Motion feature enhancement computing: A Feature Enhancement Module (FEM) is embedded in the YOLOv4 backbone network, using multi-scale convolution kernels (3 × 3/5 × 5) to extract fine-grained features of serving actions (e.g., wrist rotation angle and knee flexion). Combined with a Feature Fusion Module (FFM) in PANet’s convolutional layers, it realizes weighted fusion of high- and low-dimensional features, improving feature extraction accuracy by 23%. Health big data modeling: For multi-source health data (heart rate, swing frequency, electromyography) collected by wearable devices, a temporal interpolation algorithm is used to handle missing values. Principal Component Analysis (PCA) reduces 12-dimensional raw features to 5 core indicators, increasing data processing efficiency by 40%. Real-time decision support engine: An action-health correlation model is built, using random forest to mine potential associations between serving parameters (e.g., toss height) and health indicators (e.g., heart rate variability), generating personalized training recommendations with a decision response delay within 150 ms. Experimental validation based on a self-built tennis serving dataset (8600 action videos) and health monitoring data of 120 athletes shows: The framework achieves 97.6% accuracy in serving action recognition (8.3% higher than original YOLOv4), an F1-score of 0.91 for action-health correlation analysis, and 99.2% system stability (72-h fault-free operation). This research not only provides a computational solution for quantitative analysis in athlete training but also promotes the integration of computer vision and health big data in sports training engineering.