Enhanced Adjacency Matrix-Based Lightweight Graph Convolution Network for Action Recognition.

Area of Science:

• Computer Vision
• Machine Learning
• Artificial Intelligence

Background:

• Graph convolutional networks (GCNs) are effective for skeleton-based human action recognition.
• Existing GCN models face challenges in model construction and feature fusion.

Purpose of the Study:

• To propose an enhanced adjacency matrix-based graph convolutional network with a combinatorial attention mechanism (CA-EAMGCN).
• To improve skeleton-based human action recognition accuracy and model efficiency.

Main Methods:

• Constructing an enhanced adjacency matrix to expand the perceptive field of global node features.
• Designing a feature selection fusion module (FSFM) for optimal multi-feature fusion.
• Integrating spatial-temporal (ST) attention and limb attention modules (LAM) via a combinatorial attention mechanism.

Main Results:

• The proposed CA-EAMGCN model demonstrates effectiveness on large-scale datasets (NTU RGB+D 60, NTU RGB+D 120, UAV-Human).
• The model achieves a balance between lightweight design and high recognition accuracy.

Conclusions:

• The CA-EAMGCN model offers an effective solution for skeleton-based human action recognition.
• The proposed enhancements contribute to improved performance and efficiency in GCN-based action recognition systems.