Based on the existing video surveillance infrastructure, edge-side AI boxes or backend AI recognition and analysis supercomputers are deployed according to scenarios to conduct real-time collection and analysis of operating environments, operational risks, and unsafe behaviors of operators. This compensates for the insufficient safety management and control at grassroots production sites and inadequate supervision of monitoring centers, achieving full video coverage, pre-event identification and early warning of potential safety hazards.