Research On Detailed Power Tower Inspection Using Flying Robot Based On 3D Model Matching

Flying robots have been deployed into power tower inspection mission due to the low effiency and risks that traditional artificial inspection may face with. Electronical device inspection standard requires that the flying robot need to be capable of autonomous detailed inspection. Detailed inspection mainly consists of robot self-localization and target region recognition. Self-localization focus on estimation of flying robots’s pose relative to power tower, which is also the foundation of successive path planing, pose ajustment and view point selection. Target recognition aims on extraction of desired subregion of power tower from images which is basis for semantic inspection target identification. Therefore, this paper proposes a new detailed inspection method, relying on 3d wire frame model matching and point cloud model matching, which is refered to as 3D model matching based method. Firstly, for self localization, a mutli-sensor fusion based method using wire frame model is proposed, which combines GPS, IMU and vision information to precisely localize flying robot. Secondly, for target region recognition, this paper proposes a point cloud based target recognition method which can extract target region from images in accordance with semantic orders. Finally, a detailed inspection method and related design and implemention of its platform is illustrated. The experiment shows this method not only performs well for detailed inspection but also has a higher precision and robustness compared to traditional method.