Multi-Angle Fusion-Based Safety Status Analysis of Construction Workers

Area of Science:

• Computer Vision
• Construction Safety
• Artificial Intelligence

Background:

• Construction sites face significant safety challenges due to hazardous accidents.
• Existing computer vision methods for worker detection suffer from limited camera coverage and occlusion issues.
• Effective safety management is crucial but remains a dilemma for construction managers.

Purpose of the Study:

• To develop an improved computer vision framework for enhanced worker detection and trajectory estimation on construction sites.
• To address the limitations of single-camera systems and occlusion in identifying construction workers.
• To provide real-time safety status analysis and prior warning signals for accident prevention.

Main Methods:

• A multi-angle fusion approach combining SURF (Speeded Up Robust Features) algorithm.
• Integration of improved Gaussian Mixed Model (GMM) and Histogram of Oriented Gradient (HOG) with Support Vector Machines (SVM).
• Real-time worker tracking and movement trajectory estimation using Kalman filters.

Main Results:

• The proposed framework successfully identifies obscured construction workers.
• Achieved improved detection accuracy and expanded coverage compared to traditional methods.
• Demonstrated effective real-time worker tracking and accurate movement trajectory prediction.

Conclusions:

• The developed multi-angle fusion framework significantly enhances construction site safety management.
• The system provides reliable worker detection and trajectory prediction for proactive safety measures.
• This approach offers a robust solution for identifying workers and mitigating risks in complex construction environments.