After the automatic transformation and visualization upgrading of modern coal mines, the coverage rate of underground industrial video tends to be saturated, and massive video resources are generated at the same time. However, the utilization of video resources is still limited to the monitoring of site environment and equipment and the replay of accident tracing. In order to further develop and utilize the video resources of coal mine industry and improve the intelligent level of mine, combined with the actual needs of underground coal mine business, the development and application scenarios of video resources are designed. According to the requirements of fault point monitoring and mine production measurement in the transportation system of coal mine belt conveyor, using the vision feature modeling extraction technology based on monocular vision and the vision feature modeling extraction technology based on multiocular vision, the scene of foreign body identification, coal load identification, coal volume identification, off-track identification of belt conveyor was designed. According to the monitoring requirements of fire prone area in underground coal mine, the scene of regional smoke fire identification is designed. According to the operation requirements of the underground coal mine operators’ behavior supervision, the identification scene of operators’ illegal behavior is designed. The in-depth development and utilization of the video of coal mine industry based on AI-based video analysis function can eliminate a lot of garbage data, retain effective data, and relieve the transmission pressure brought by the video data flow to the mine backbone industrial ring network. Through computer intelligent identification, analysis and judgment, it can replace part of the functions of inspection personnel, reduce the labor intensity of underground workers, and improve the safety production capacity of mine without adding additional sensing equipment.