Abstract: To ensure the safe production of underground metal mines and effectively reduce the "three types of violations" (command violations，operation violations，and labor discipline violations)，an intelli⁃ gent video monitoring system for underground metal mines is researched and developed to realize real-time monitoring and intelligent analysis of mine production.Functions such as high-definition video acquisition， edge or server-side intelligent analysis，3D visualization，electronic fencing，and early warning for emer⁃ gency events are organically integrated，and an architecture adapted to high humidity，high dust levels，and blasting impacts is constructed. After analysis and evaluation，target recognition based on convolutional neu⁃ ral networks ，behavior analysis based on temporal models，and a model update strategy combining transfer learning and online fine-tuning are proposed to improve recognition accuracy and reduce the false alarm rate.In view of the impact of the underground environment on image quality，equipment stability，and net⁃ work transmission，countermeasures such as equipment selection，monitoring point optimization，and en⁃ hanced maintenance are put forward.The research results show that the system has significant application value in improving the efficiency of identifying violation behaviors，shortening the response time for emer⁃ gency events，reducing the labor intensity of patrol inspections，and enhancing mine safety management and production efficiency. It also provides a feasible path and implementation suggestions for the construction of smart mines.

Key words: underground metal mine, intelligent video monitoring system, artificial intelligence