Related Experiment Video
Updated: Jul 2, 2025
06:37
Author Spotlight: Addressing Technical and Subjective Challenges in Measuring Classroom Attention
Published on: December 15, 2023
3.8K
Skeleton-Based Activity Recognition for Process-Based Quality Control of Concealed Work via Spatial-Temporal Graph
Lei Xiao1, Xincong Yang2, Tian Peng3
1Department of Building and Real Estate, The Hong Kong Polytechnic University, Hong Kong, China.
Sensors (Basel, Switzerland)
|February 24, 2024
Summary
This study introduces a computer vision framework using Spatial-Temporal Graph Convolutional Networks (ST-GCNs) for real-time construction quality control. The model accurately recognizes plastering activities and their order, enabling detection of missing or misplaced steps.
Area of Science:
- Construction Engineering and Management
- Computer Vision
- Artificial Intelligence
Background:
- Computer vision (CV) automates construction site monitoring but is underutilized for process-based quality control, particularly for concealed works.
- Current methods lack real-time, automated analysis of construction sequences and quality adherence.
- There is a need for advanced techniques to ensure quality in hidden construction processes.
Purpose of the Study:
- To develop and validate a framework for process-based quality control in construction using Spatial-Temporal Graph Convolutional Networks (ST-GCNs).
- To enable automated recognition of construction activities and their temporal order for quality assessment.
- To address the gap in applying CV for real-time quality control of concealed construction works.
Main Methods:
- A framework utilizing Spatial-Temporal Graph Convolutional Networks (ST-GCNs) was developed.
- An on-site plastering work video dataset was collected for experimental validation.
- The ST-GCN model was trained to recognize four primary plastering activities and their sequence.
Main Results:
- The ST-GCN model achieved 99.48% accuracy in recognizing plastering activities on the validation set.
- The model successfully identified correct activity sequences, missing activities (e.g., fiberglass mesh covering), and incorrect activity orders in test videos.
- Activity order recognition was effective, allowing for convenient judgment of process integrity.
Conclusions:
- The developed ST-GCN framework offers a promising approach for active, real-time, process-based quality control in construction.
- This technology can significantly enhance quality assurance for concealed works by automating sequence and activity verification.
- The study demonstrates the potential of advanced CV techniques to improve construction process monitoring and defect detection.
