Arm Movement Analysis Technology of Wushu Competition Image Based on Deep Learning

Area of Science:

• Computer Vision
• Sports Science
• Artificial Intelligence

Background:

• Recognizing action poses in martial arts requires analyzing sequential movements, which single frames cannot capture.
• Existing methods may lack the temporal dynamics crucial for accurate pose recognition in dynamic sports.

Purpose of the Study:

• To develop a deep learning-based technology for analyzing arm movements in martial arts competitions.
• To improve the accuracy of action pose recognition by incorporating spatiotemporal features.

Main Methods:

• Extracted arm motion features from bone sequences, combining them with RGB spatial and depth map data.
• Utilized a deep learning model with dual frame rate channels (slow and fast) to analyze 16-frame video samples.
• Employed a softmax classifier for action category classification.

Main Results:

• The proposed arm motion analysis technology achieved an accuracy rate of 95.477%.
• A recall rate of 92.948% was obtained, indicating strong performance in action recognition.
• The method effectively captured spatiotemporal features for robust motion analysis.

Conclusions:

• Deep learning effectively analyzes martial arts arm movements by integrating bone, RGB, and depth data.
• The spatiotemporal feature extraction method significantly enhances action pose recognition accuracy.
• This technology demonstrates good motion analysis performance for martial arts competitions.