| Power transmission lines undertake the important task of transmitting electrical energy and require regular inspections and maintenance.With the development of Unmanned Aerial Vehicle technology,intelligent UAV autonomous inspection system is one of the current key research directions.The route planning of the existing UAV inspection is basically designed based on the GPS information of the waypoint.The GPS signal in some areas is weak,and the use conditions are limited.This is the main bottleneck hindering the widespread application of UAV power inspection.In view of the situational awareness of power transmission lines in complex environments,this paper uses multi-rotor drones as a platform to build an UAV power inspection system with autonomous flight control capabilities and defect detection capabilities through the comprehensive use of drone technology,visual positioning technology and deep learning technology,named autonomous inspection system of UAV based on deep vision,which realizes real-time inspection of insulator hazards with greater harm and obvious characteristics.Main tasks as follows:Firstly,a deep vision positioning subsystem is proposed,using a target detection algorithm based on deep learning,and using the projection relationship of monocular vision to perform spatial positioning on detected objects of known size,without the need for positioning equipment or visual mapping.The nonlinear monocular vision positioning model was studied to correct the effects of camera lens distortion and optical axis calibration errors,and improve positioning accuracy.Comprehensive consideration of detection accuracy and speed,designing a lightweight object detection network MobileNetV1-YOLOv3 suitable for portable intelligent platforms,establish the connection between the detection results and the visual positioning model,convert the extracted visual information into a spatial location and detect defects.Secondly,in view of the problem of lagging in the output of visual positioning information,a Kalman filtering algorithm fused with multi-sensor data is proposed.The airborne sensor data is used to correct and compensate the visual positioning information to achieve real-time positioning of the UAV.Based on the Android platform and DJI Mobile SDK,the ground station control center was developed,and the implementation methods of core function modules such as information interaction of hardware devices and virtualStick mode control of the drone were specifically discussed.Finally,the overall performance of the system in this paper is tested in the field environment,the system operation process is designed and the positioning error of the system is quantitatively analyzed through comparative experiments.Then the actual inspection of the insulator is conducted and a complete inspection process is designed for the 220 kV power pole.It shows that the positioning accuracy of the system in this paper is better than that of GPS equipment and can realize the real-time inspection of insulators by UAVs. |
