Abstract:At present,the patrol point setting for industrial scene patrol robots is manual setting of personnel.Considering that the number of patrol points is usually large in the complex environment of the site,setting tasks consumes a lot of manpower and time,and once the equipment of the patrol points changes,manually adding or deleting patrol points is needed,a robot patrol method based on automatic visual scanning to generate patrol tasks was proposed .Using the video stream of the robot scanning environment,all pictures were sent to YOLOv4 target detection network for analysis and screening,and the SIFT matching method was used to remove duplicate targets,then patrol points could automatically generated for complex field equipment.This method was applied in the robot patrol projects of tunnel cable.Compared with manual setting mode,it can improve the efficiency over 90%.