The Design Of Vision System For Power Tower Climbing Robot

With the rapid development of the power industry,the need to replace robots for intelligent inspections is increasing,and special robots for power operations have emerged.As one of the special intelligent inspection robot products,the power tower climbing robot’s main function is to climb the power tower and complete the related inspection work.Achieving the automation of power tower climbing robots is one of the key goals in the era of intelligent inspection.It is of great significance for improving inspection efficiency,reducing production costs,accelerating the intelligentization of power special operations robots,and ensuring national energy security.This paper designed a vision system for the power tower climbing robot which has the functions of positioning climbing foot-nails and recognizing posture of the safety lock,which provides an idea for improving the level of automation of the power tower climbing robot.The main content of the full text is as follows:(1)Summarized the research status of existing power tower climbing robots,analyze the key technologies of robot vision system,and research the excellent results in the field of target detection and gesture recognition.Combined with the robot mechanical system and application requirements,a suitable vision system configuration method was selected.The visual system software and hardware framework is designed for the robot climbing tower positioning function and anti-fall protection function,and the spatial positioning description method is analyzed.(2)To realize the function that the robot locates the foot nail to climb the tower,the deep learning algorithm is adopted in the vision system to identify and locate the foot nail.Mobilenet’s deep separable ideas are used to optimize the efficient target detection model YOLOv3,and the prior frame clustering method is optimized.The model was trained through a data set composed of a self-made tower model and outdoor real scenes,and the foot nail recognition was realized.Then the depth characteristics of the camera are analyzed,and the recognition results are combined with the depth information of the Intel Realsense D435 camera to realize the positioning of the foot nails with high accuracy.(3)To realize the anti-fall protection function,the point cloud registration method is adopted in the vision system to identify the spatial attitude of the safety lock.Through the Intel Realsense D435,the scene point cloud was acquired and processed such as filtering and segmentation,and a template point cloud database was constructed based on the threedimensional model.The optimal posture template of the security lock is obtained by calculating the feature matching degree between the scene point cloud and the template point cloud,and then the point cloud registration is performed based on SAC-IA and ICP algorithm to realize posture recognition and optimization.It is obtained through experiments that the gesture recognition result of the safety lock has high accuracy.(4)Completed the development of the vision system,and applied the identification and positioning of the power tower studs and the safety lock posture recognition in the field environment.It is of great significance for accelerating the intelligentization and automation of power tower climbing robots,and provides new ideas for improving the level of electric power intelligent inspection.