An Intelligent and Fast Dance Action Recognition Model Using Two-Dimensional Convolution Network Method

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Action recognition in computer vision is challenging.
• Dance movement recognition requires specialized methods.
• Existing models may be computationally intensive.

Purpose of the Study:

• To propose a dance movement recognition method using a deep learning (DL) network.
• To improve accuracy and efficiency compared to existing algorithms.
• To leverage spatio-temporal features for dance analysis.

Main Methods:

• Utilized MobileNet as the backbone network.
• Integrated time-domain feature transfer between convolutional layers.
• Employed clustering for human detection prior frames.
• Developed a DL network for dance movement recognition.

Main Results:

• The proposed algorithm outperformed Incision v3 in F1 score by 9.87%.
• Achieved higher identification accuracy than traditional CNN algorithms (6.51% and 10.76%).
• Demonstrated reduced running time and improved dance recognition accuracy.

Conclusions:

• The DL-based method effectively recognizes dance movements.
• The approach offers a balance between accuracy and computational efficiency.
• Provides a valuable reference for future research in dance action recognition.