Dance Fitness Action Recognition Method Based on Contour Image Spatial Frequency Domain Features and Few-Shot...

Area of Science:

• Artificial Intelligence
• Computer Vision
• Biomedical Engineering

Background:

• Few-Shot learning models have advanced rapidly, achieving success in diverse AI applications like image recognition and natural language processing.
• Identifying dance fitness movements is crucial for personalized training and health monitoring.

Purpose of the Study:

• To develop an effective Few-Shot learning method for dance fitness movement identification.
• To leverage contour image spatial frequency domain features for enhanced action recognition.

Main Methods:

• A novel Few-Shot learning approach utilizing contrastive average loss is proposed.
• The method incorporates an improved loss function and normalization for model representativeness.
• A feature extraction algorithm combining improved Local Binary Patterns (LBP) and Histogram of Oriented Gradients (HOG) is developed.

Main Results:

• The proposed algorithm achieved high recognition accuracies: 93.10% for walking, 90.30% for opening hands, and 92.70% for running.
• Experimental results demonstrate the effectiveness of the fusion feature algorithm in dance fitness action recognition.

Conclusions:

• The developed Few-Shot learning method, combined with fused LBP and HOG features, shows significant promise for accurate dance fitness movement identification.
• This approach contributes to advancing AI applications in the health and fitness domain.