A Spatial-Temporal Multi-Feature Network (STMF-Net) for Skeleton-Based Construction Worker Action Recognition.

Area of Science:

• Construction Management
• Artificial Intelligence
• Human-Computer Interaction

Background:

• Monitoring construction worker productivity, health, and safety is a global concern.
• Existing methods for action recognition often rely on single-stream data, limiting feature capture.
• Advances in pose estimation and deep learning offer potential for automated worker action assessment.

Purpose of the Study:

• To develop an efficient method for continuous monitoring and timely action recognition of construction workers.
• To address the limitations of single-stream data in previous action recognition studies.
• To propose a novel deep learning model for enhanced worker action assessment.

Main Methods:

• Utilized 3D skeleton and joint trajectory data from construction sites.
• Developed a Spatial-Temporal Multi-Feature Network (STMF-Net).
• Incorporated six types of 3D skeleton-based features for comprehensive movement capture.

Main Results:

• The STMF-Net achieved an accuracy of 79.36% in recognizing construction worker actions.
• The model effectively captures and processes multi-feature 3D skeleton data.
• Demonstrated the capability of deep learning for robust workforce action assessment.

Conclusions:

• The proposed STMF-Net enhances the ability to monitor and recognize construction worker actions.
• This technology has the potential to improve construction industry management models.
• Ultimately, the research aims to boost worker health and work efficiency through advanced monitoring.