Publications / CCC 2025 - Zadar, Croatia

COMPUTER VISION AND LARGE LANGUAGE MODEL-BASED SAFETY MANAGEMENT FOR CONSTRUCTION PROJECT SITES

Bin Tang, Hanbin Luo
Pages 249-257 (CCC 2025 - Zadar, Croatia, ISBN 978-1-7643710-0-1, ISSN 2413-5844)
Abstract:

Construction sites are inherently hazardous environments where accidents and fatalities constitute a significant global concern. Despite continuous efforts to enhance safety, the incidence of workplace accidents within the construction industry remains elevated. This study introduces HUST AI Box, an AI and computer vision-based safety management system, enhanced by the integration of DeepSeek, a large language model (LLM), to provide advanced natural language processing (NLP) capabilities for real-time decision support, knowledge management, and safety optimization. The integration of DeepSeek enables the system to process unstructured text data and provide actionable insights, significantly enhancing the safety management framework. We demonstrate the localized deployment of DeepSeek within the HUST AI Box system, ensuring robust performance and real-time responsiveness in dynamic construction environments. The results suggest that the combined system significantly enhances safety management by automating the detection of unsafe behaviours, improving the efficiency and precision of face recognition, and leveraging DeepSeek for real-time decision support and knowledge management. This system is versatile and can be applied in various other contexts beyond construction sites.

Keywords: Artificial intelligence (AI), Computer vision, Safety management, Construction sites, Large language model (LLM), Localized deployment