Graph Convolutional Networks for multi-modal robotic martial arts leg pose recognition

Area of Science:

• Computer Vision
• Machine Learning
• Sports Science

Background:

• Accurate martial arts leg pose recognition is crucial for sports analytics, rehabilitation, and HCI.
• Traditional models struggle with complex spatial-temporal dynamics in martial arts movements.

Purpose of the Study:

• To develop a robust model for accurate martial arts leg pose recognition.
• To overcome limitations of existing sequential and convolutional approaches.

Main Methods:

• Proposed PoseGCN, a Graph Convolutional Network (GCN) model.
• Integrated spatial-temporal graph encoding, action-specific attention, and self-supervised learning.
• Evaluated on Kinetics-700, Human3.6M, NTU RGB+D, and UTD-MHAD datasets.

Main Results:

• PoseGCN achieved state-of-the-art accuracy and F1 scores across benchmark datasets.
• Demonstrated superior performance in capturing complex spatial-temporal dependencies.
• Showcased excellent generalization capabilities across diverse datasets.

Conclusions:

• PoseGCN offers a robust solution for precise martial arts action recognition.
• The model effectively captures fine-grained pose details and temporal progression.
• Paves the way for advancements in multi-modal pose analysis.