Modeling of SPM-GRU ping-pong ball trajectory prediction incorporating YOLOv4-Tiny algorithm

Area of Science:

• Computer Vision
• Deep Learning
• Sports Analytics

Background:

• Accurate table tennis ball tracking is crucial for performance analysis.
• Existing methods face challenges in real-time accuracy and varying lighting conditions.

Purpose of the Study:

• To improve the accuracy and real-time tracking of table tennis ball trajectories.
• To develop an advanced algorithm combining physical models and deep learning for motion analysis.

Main Methods:

• Utilized a micro-miniature fourth-generation real-time target detection algorithm with a gated loop unit.
• Integrated physical models with deep learning for enhanced table tennis ball motion analysis.
• Conducted comparative experiments against traditional target detection and trajectory prediction models.

Main Results:

• The improved algorithm achieved an 86.74% multi-target recognition accuracy (22.36% increase) and 89.12% accuracy in low light.
• Processing efficiency reached 85 frames/s.
• Trajectory prediction errors were significantly reduced to 4.5 mm, 25.3 mm, and 35.58 mm compared to traditional models.

Conclusions:

• The developed model significantly improves table tennis ball position and trajectory tracking accuracy.
• Offers practical value for training and strategy in table tennis.
• Provides a reference for applying similar techniques in other sports analytics.