Deep learning-based system for prediction of work at height in construction site

Area of Science:

• Occupational Safety and Health
• Artificial Intelligence in Construction
• Wearable Sensor Technology

Background:

• Falling from height (FFH) is a primary cause of construction site injuries and fatalities.
• Technological advancements are crucial for mitigating FFH risks.
• Current safety measures may lack real-time predictive capabilities for working at heights.

Purpose of the Study:

• To forecast if a construction laborer is operating at an elevated position using sensor data.
• To apply deep learning techniques for real-time safety monitoring.
• To bridge information gaps in predicting workers' conditions at heights.

Main Methods:

• Utilized accelerometer, gyroscope, and pressure data from workers.
• Applied deep learning models: Deep Neural Network (DNN), Convolutional Neural Network (CNN), and Long Short-Term Memory (LSTM).
• Conducted 45 analyses varying window sizes (e.g., 1-s, 5-s) and overlap rates (e.g., 75%).

Main Results:

• The DNN model achieved 94.6% accuracy with a 1-s window and 75% overlap.
• The CNN model achieved 94.9% accuracy with a 5-s window and 75% overlap.
• Both models demonstrated high predictive accuracy without complex calculations.

Conclusions:

• Deep learning models can efficiently predict workers' elevated working conditions.
• This predictive capability can enable timely implementation of safety measures.
• The study supports integrating advanced technology to reduce FFH incidents and improve construction safety.