Abstract: To address the issues of low efficiency in manual inspection, delayed hazard identification, poor performance of intelligent recognition algorithms, and incomplete systems in traditional construction safety management, this study proposes an intelligent safety management method integrating a structured knowledge base with video monitoring. First, through the analysis of industry safety risk cases and rule-based descriptions, a knowledge base containing 12 categories and 876 typical safety risk sources was established. Subsequently, a data screening and quality control method for construction site video monitoring was developed, leading to the creation of an object detection dataset and a construction element detection model based on the YOLOv8 algorithm. By leveraging the safety risk source knowledge base, the detected construction elements are associated with corresponding risk sources to enable intelligent identification of typical safety hazards. Practical applications in two typical scenarios – dynamic monitoring of personnel intrusion into hazardous areas and real-time supervision of hot work operations – demonstrate significant improvements: compared with traditional management methods, the rapid detection and rectification rate of hazards reached 92%, while the intervention success rate for risky behaviors increased by 22%. This approach effectively enhances the efficiency and intelligent level of safety management at construction sites.