Skeleton-Based Human Action Recognition With Global Context-Aware Attention LSTM Networks.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Human action recognition from 3D skeleton data is a significant research area.
• Long Short-Term Memory (LSTM) networks excel at modeling sequential data but lack explicit attention mechanisms.
• Irrelevant skeletal joints can introduce noise, hindering action recognition performance.

Purpose of the Study:

• To develop an attention-based LSTM network for skeleton-based human action recognition.
• To enhance the model's ability to focus on informative skeletal joints.
• To improve the overall accuracy and robustness of action recognition systems.

Main Methods:

• Proposed a global context-aware attention LSTM (GCALSTM) incorporating a global context memory cell.
• Introduced a recurrent attention mechanism for progressive enhancement of attention capabilities.
• Developed a two-stream framework utilizing both coarse-grained and fine-grained attention.

Main Results:

• The GCALSTM model demonstrated superior performance in selectively focusing on informative joints.
• The recurrent attention mechanism progressively improved the network's attention effectiveness.
• The two-stream framework enhanced the model's ability to capture multi-level spatial-temporal features.

Conclusions:

• The proposed global context-aware attention LSTM significantly advances skeleton-based human action recognition.
• The method achieves state-of-the-art results on five challenging benchmark datasets.
• This work offers a more effective approach to handling noisy and redundant information in skeletal data.