Author(s)

Xiangyi Wang

Corresponding Author

Xiangyi Wang

Abstract

This paper focuses on the difficult problem of safety management in the construction site, aiming at realizing accurate identification and real-time intervention of safety behaviors with the help of artificial intelligence (AI) technology. By setting up an experimental environment, equipment was deployed at the construction site to collect 8000 image data covering a variety of safe and unsafe behaviors, and the CNN-LSTM hybrid model was trained. After testing, the model performs well in behavior recognition such as helmet wearing, equipment operation and seat belt wearing, with the accuracy of 96%, 94% and 95% respectively. The average processing time of a single image is 0.04 seconds, and the average response time of the system is 0.18 seconds. The research shows that the AI-based system is highly effective and real-time in identifying and real-time intervening safety behaviors in the construction site, which provides an innovative and feasible scheme for improving the safety management level of the construction site, but the accuracy of identification in complex environment still needs to be improved.

Keywords

Artificial Intelligence; Construction Site; Safety Behavior Identification; Real-Time Intervention; CNN-LSTM Model