Deep learning neural network-assisted badminton movement recognition and physical fitness training optimization

Area of Science:

• Computer Vision
• Robotics
• Sports Analytics

Background:

• Accurate recognition of badminton movement trajectory is crucial for robotic systems.
• Existing methods struggle with the small target detection of flying shuttlecocks in video streams.

Purpose of the Study:

• To improve the accuracy of badminton movement trajectory recognition.
• To enhance the detection and tracking capabilities for flying shuttlecocks in badminton robots.

Main Methods:

• Developed an improved deep learning one-stage detection network, Tiny YOLOv2, by modifying the loss function and network structure.
• Integrated an attention mechanism into a convolutional neural network for trajectory identification.
• Combined the enhanced Tiny YOLOv2 with the Unscented Kalman Filter for trajectory prediction.

Main Results:

• The proposed method achieved an average accuracy of 91.40% and a recall rate of 84.60% for tracking badminton trajectories.
• In various scenarios, the system demonstrated average precision of 96.7%, recall of 95.7%, and a frame rate of 29.2 frames/second.
• Outperformed existing algorithms in tracking and predicting badminton trajectories.

Conclusions:

• The improved algorithm significantly enhances badminton trajectory recognition.
• This method provides robust support for badminton movement analysis and robotic applications.